NBER WORKING PAPER SERIES

THE DIFFUSION OF DEVELOPMENT

Enrico SpolaoreRomain Wacziarg

Working Paper 12153http://www.nber.org/papers/w12153

NATIONAL BUREAU OF ECONOMIC RESEARCH1050 Massachusetts Avenue

Cambridge, MA 02138March 2006

Spolaore: Department of Economics, Tufts University, 315 Braker Hall, 8 Upper Campus Rd., Medford, MA02155-6722, enrico.spolaore@tufts.edu. Wacziarg: Graduate School of Business, Stanford University,Stanford CA 94305-5015, USA, wacziarg@gsb.stanford.edu. We thank Alberto Alesina, Robert Barro,Francesco Caselli, Steven Durlauf, Jim Fearon, Oded Galor, Yannis Ioannides, Pete Klenow, AndrosKourtellos, Ed Kutsoati, Peter Lorentzen, Paolo Mauro, Sharun Mukand, Gérard Roland, AntonioSpilimbergo, Chih Ming Tan, David Weil, Bruce Weinberg, Ivo Welch, and seminar participants at BostonCollege, Fundação Getulio Vargas, Harvard University, the IMF, INSEAD, London Business School,Northwestern University, Stanford University, Tufts University, UC Berkeley, UC San Diego, the Universityof British Columbia, the University of Connecticut and the University of Wisconsin-Madison for comments.We also thank Robert Barro and Jim Fearon for providing some data. The views expressed herein are thoseof the author(s) and do not necessarily reflect the views of the National Bureau of Economic Research.

©2006 by Enrico Spolaore and Romain Wacziarg. All rights reserved. Short sections of text, not to exceedtwo paragraphs, may be quoted without explicit permission provided that full credit, including © notice, isgiven to the source.

The Diffusion of DevelopmentEnrico Spolaore and Romain WacziargNBER Working Paper No. 12153March 2006JEL No. O11, O57

ABSTRACT

This paper studies the barriers to the diffusion of development across countries over the verylong-run. We find that genetic distance, a measure associated with the amount of time elapsed sincetwo populations’ last common ancestors, bears a statistically and economically significant correlationwith pairwise income differences, even when controlling for various measures of geographicalisolation, and other cultural, climatic and historical difference measures. These results hold not onlyfor contemporary income differences, but also for income differences measured since 1500 and forincome differences within Europe. We uncover similar patterns of coefficients for the proximatedeterminants of income differences, particularly for differences in human capital and institutions.The paper discusses the economic mechanisms that are consistent with these facts. We present aframework in which differences in human characteristics transmitted across generations - includingculturally transmitted characteristics - can affect income differences by creating barriers to thediffusion of innovations, even when they have no direct effect on productivity. The empiricalevidence over time and space is consistent with this "barriers" interpretation.

Enrico SpolaoreDepartment of EconomicsTufts University315 Braker Hall8 Upper Campus Rd.Medford, MA 02155-6722enrico.spolaore@tufts.edu

Romain WacziargStanford Graduate School of Business518 Memorial WayStanford UniversityStanford, CA 94305-5015and NBERwacziarg@gsb.stanford.edu

1 Introduction

What explains the vast differences in income per capita that are observed across countries? Why

are these differences so persistent over time? In this paper, we argue that barriers to the diffusion of

development prevent poor countries from adopting economic practices, institutions and technologies

that make countries rich. We argue that these barriers are not only geographic, but also human.

We propose and test the hypothesis that cross-country differences in human characteristics that are

transmitted with variations from parents to children create barriers to the diffusion of development.

These vertically transmitted characteristics include cultural features, such as language and habits,

among other characteristics of human populations.

In recent years a large empirical literature has explored the determinants of income levels using

cross-country regressions, in which the level of development, measured by income per capita, is

regressed on a set of explanatory variables.1 In this paper, we depart from this usual methodology.

We are not primarily concerned with the factors that make countries rich or poor. Instead, we are

concerned with the barriers that prevent poor countries from adopting better income determinants,

whatever they are. To do so, we use income differences between pairs of countries as our depen-

dent variable, and various measures of distance as regressors. Our approach allows us to consider

measures of distance between countries across several dimensions, and to investigate whether those

distances play a role as barriers to the diffusion of development.2

For the first time, we document and discuss the relationship between genetic distance and

differences in income per capita across countries. We use genetic distance between populations as

a measure of their degree of similarity in vertically transmitted characteristics (VTCs). Measures

1Recent contributions to this literature include Hall and Jones (1999), Acemoglu, Johnson and Robinson (2001),

Easterly and Levine (2003), Alcalá and Ciccone (2004), Rodrik, Subramanian and Trebbi (2004), Glaeser, La Porta,

Lopez-de-Silanes and Shleifer (2004) among others.

2There is a voluminous literature on cross-country income convergence, dating back to Baumol (1986). In the

neoclassical literature, convergence occurs because the marginal return to capital is higher in countries farther from

their steady-state, which depends, among other things, on the level of technology (the "A" parameter). In contrast,

we seek to shed light on the factors that prevent or facilitate the diffusion of productivity-enhancing innovations

across countries. In this respect, our paper is closer to the approach in Barro and Sala-i-Martin (1997), where

technological diffusion drives convergence. Policy-induced constraints on the diffusion of technology are analyzed by

Parente and Prescott (1994, 2002). Policy experimentation and imitation across neighbors are studied by Mukand

and Rodrik (2005). Unlike these contributions, we consider more broadly the barriers to the diffusion of technological

and institutional characteristics in the very long run.

1

of genetic distance between populations are based on aggregated differences in allele frequencies

for various loci on a chromosome. In this paper we use measures of FST distance, also known

as coancestor coefficients. FST distances, like most measures of genetic diversity, are based on

indices of heterozygosity, the probability that two genes at a given locus, selected at random from

the relevant populations, will be different (heterozygous). Since most genetic differences tend to

accumulate at a regular pace over time, as in a kind of molecular clock, genetic distance is closely

linked to the time since two populations’ last common ancestors - that is, the time since two

populations were in fact the same population. Hence, genetic distance can be used to determine

paths of genealogical relatedness of different populations over time (phylogenetic trees).3

The main findings of this paper are fourfold. First, measures of genetic distance between

populations bear a statistically and economically significant effect on differences in income per

capita, even when controlling for various measures of geographical isolation, and other cultural,

climatic and historical difference measures. Second, the effect of genetic distance holds not only for

contemporary income differences, but also for income differences measured since 1500. While the

effect is always large, positive, and significant, the magnitude of the effect has varied over time in

an interesting way. The effect declined from 1500 to 1820, went up dramatically, peaked at the time

of the Industrial Revolution, and steadily declined afterwards. Third, the effect of genetic distance

on income differences is larger for countries that are geographically closer. Finally, the effect of

genetic distance holds not only for contemporary and historical worldwide income differences, but

also for income differences within Europe. The magnitude of the effect of genetic distance is larger

within European countries than across countries from all continents.

In a nutshell, the correlation between genetic distance and income differences is extremely robust

over time and space, but also presents important variations over those dimensions. These variations

over time and space provide valuable clues about the economic interpretation of the effect.

What is the economic meaning of this effect? One possibility is that this correlation may

just reflect the impact of variables affecting both genetic distance and income differences. If that

were the case, controlling for those variables would eliminate the effect of genetic distance on

3Our main source for genetic distances between human populations is Cavalli-Sforza, Menozzi and Piazza (1994).

The classical reference on evolutionary rates at the molecular level is Kimura (1968). A recent textbook reference on

human population genetics is Jobling, Hurles and Tyler-Smith (2004). For a nontechnical discussion of these concepts

see Dawkins (2004).

2

income difference. We control for a large number of reasonable suspects (geographical and climatic

differences, measures of geographical isolation, etc.). In particular, we control for geography and

region-specific differences that may impede the diffusion of development, as emphasized by Jared

Diamond in his influential book Guns, Germs and Steel (1997).4 We find that these geographical

and regional variables often do have an effect on income differences, but that their inclusion does

not eliminate the effect of genetic distance as an independent explanatory variable. Moreover, the

effect of genetic distance on income differences holds within Europe, where geographic differences

are much smaller.

Our empirical analysis opens the door to a causal interpretation of the relationship between ge-

netic distance and income differences. What mechanisms can explain a causal link? It is important

to stress that a link from genetic distance to income differences is not evidence that the mechanisms

themselves are genetic. On average populations that are more genetically distant have had more

time to diverge in a broad variety of characteristics transmitted intergenerationally. These include

characteristics that are passed on genetically, through DNA, but also some that are passed on

non-genetically, i.e. culturally.5 As long as these cultural characteristics are transmitted to younger

generations from genetically related individuals, they will be correlated with genetic distance. Lan-

guage is an obvious example. While humans are genetically predisposed to learn some language,

there is no gene for speaking Japanese or Italian. However, people who speak the same language

tend to be closely related genetically because most children learn their language from their parents.

Moreover, since languages (and other deep cultural characteristics) change gradually over time,

people who speak more similar languages also tend to be closer to each other genealogically.6

Therefore, one should not view genetic distance as an exclusive measure of distance in DNA-

transmitted characteristics. It is more appropriate to interpret genetic distance as a general metric

4See also Olsson and Hibbs (2005).

5By vertical cultural transmission we mean any transmission of characteristics from parents to children that does

not take place through DNA, such as language. Evolutionary models of cultural transmission have been developed,

by Cavalli-Sforza and Feldman (1981) and Boyd and Richerson (1985). For a nontechnical discussion see Cavalli-

Sforza and Cavalli-Sforza (1995, chapter 8). Economic models of cultural transmission from parents to children have

been provided by Bisin and Verdier (2000, 2001). Galor and Moav (2003) present an innovative theory of long-term

economic growth in which a key role is played by evolutionary changes in preference parameters that are genetically

transmitted across generations. For an in-depth discussion of these issues, see also Galor (2005).

6See Cavalli-Sforza, Menozzi and Piazza, 1994, pp. 96-105.

3

for genealogical distance between populations, capturing overall average differences not only in

genetically transmitted features but also in culturally transmitted characteristics. In this paper we

will define vertically transmitted characteristics (VTCs) to be all characteristics passed on from

parents to children, whether through DNA or culturally. If we take this broader perspective, we can

interpret the effect of genetic distance on income differences as evidence of an important role for

vertically-transmitted characteristics, reflecting divergent historical paths of different populations

over the long run.7

Rather than addressing the "nature versus nurture" debate, which is beyond the scope of our

analysis, we interpret our findings as evidence for the economic importance of long-term divergence

in VTCs of different populations: the diffusion of development is impeded by barriers arising from

differences in VTCs. That said, it is also true that we find clues pointing to cultural transmission

rather than purely genetic transmission as a likely mechanism behind our results. For instance, as

we already mentioned, we find large effects of genetic distance on income differences within Europe.

That is, genetic distance explains income differences between populations that are geographically

close, have shared very similar environments, and have had a very short time to diverge genetically

(in many cases, less than a few thousand years). Since cultural change is much faster than genetic

change, and most genetic change, especially in the short-run, is neutral (i.e. unrelated to natural

selection), our findings are consistent with cultural transmission as a key mechanism explaining

persistent income differences.8

While we do not wish to push the distinction between genetic and cultural transmission too

far, we do stress a different distinction. That is the distinction between a direct effect and a

barrier effect. VTCs have a direct effect if they enter directly into the production function - say, by

improving total factor productivity. An example would be the transmission of a more productive

work ethic from parents to children. By contrast, a barrier effect occurs if different characteristics

between populations prevent or reduce the diffusion of productivity-enhancing innovations (more

productive technology, institutions, etc.). For example, differences in language may have no direct

7Moreover, as we briefly discuss in Section 2.2, any sharp distinction between "genetic" and "cultural" charac-

teristics may be misleading, since the economic impact of genetic and cultural characteristics is likely to depend on

their combination and interaction.

8The view that cultural transmission trumps genetic transmission in explaining differences within human popu-

lations is standard among geneticists and anthropologists. For nontechnical discussions of these issues, see Diamond

(1992, 1997), Cavalli-Sforza and Cavalli-Sforza (1995), Olson (2002) and Richerson and Boyd (2004).

4

bearing on productivity but may act as obstacles to the introduction of innovations arising from

populations with different languages. Another way of stating this idea is to say that differences in

VTCs are obstacles to the horizontal diffusion of development.9 Hence, in principle, genetic distance

may explain income differences because of direct effects (some populations have more productive

VTCs than others), barrier effects (different VTCs prevent the horizontal diffusion of innovations),

or both. It is worth noting that either effect would be sufficient to account for the correlation we

document.10

Generally, a precise decomposition of the two effects is conceptually and empirically difficult, as

some VTCs may have both direct and barrier effects. However, our data provide clear indications

that VTCs act at least in part as barriers to the diffusion of development. First, there exists a

negative interaction between genetic distance and geographical distance. That is, we find that

genetic distance has a bigger effect on income differences for country pairs that are geographically

close. This result is consistent with a simple model in which geographical and genetic distance are

both barriers to the diffusion of innovations. The intuition is straightforward: if genetic distance

acts as a barrier, it matters more for countries that are nearby, and face lower geographical barriers

to exchange with each other, while it is less important for countries that are far away, and would

learn little from each other anyway due to geographic distance.11 Second, there is a pattern in the

effect of genetic distance on income differences over time, from 1500 to today: while the effect of

genetic distance is always large, positive and significant, it varies over time in an interesting way.

The effect declined from 1500 to 1820, spiked up and peaked in 1870, and steadily declined again

afterwards. This is consistent with the interpretation of genetic distance as related to barriers

to the diffusion of innovations. The effect of barriers should peak when a major innovation is

introduced and initially adopted only by the populations that are closest to the innovator (such as

the industrial revolution in the 19th century), but decline over time as the major innovation spreads

to more distant populations.

Our paper is organized as follows. In Section 2, we present a simple analytical framework to help

9 In the anthropology literature, vertical transmission takes place across (usually related) generations, while hori-

zontal transmission takes place across (possibly unrelated) groups of people belonging to the same generation.

10 In fact, Section 2.1 presents a simple model in which barriers associated with differences in neutral VTCs are

sufficient to explain a positive effect of genetic distance on income differences.

11By contrast, a simple model where VTCs and geographical characteristics directly affect the production function

would not typically imply a negative interaction term.

5

in the interpretation of our empirical work. Our simple model illustrates a) the link between genetic

distance and distance in VTCs, and b) the link between differences in VTCs and the diffusion of

innovations across populations. A key point is to show that random divergence in neutral VTCs

is sufficient to generate income differences if those VTCs are barriers to the horizontal diffusion of

innovations. An extension to non-neutral VTCs strengthens the link between genetic distance and

income differences. This section also presents and discusses a general taxonomy of the different

channels through which genetic distance may affect income differences. Section 3 discusses the

data and the empirical methodology used in this paper. Since we regress pairwise differences in

income on distance measures, we face a problem of spatial correlation, and address this estimation

issue using a new econometric methodology. Section 4 presents our empirical results: consistent

with our theoretical framework, we document that genetic distance is positively related to pairwise

differences in income per capita and in its proximate determinants. Section 5 concludes.

2 Theoretical Framework

In the first part of this section (Section 2.1) we propose a simple analytical framework to study

the diffusion of technological and institutional innovations across societies and its relationship with

genetic distance In the second part (Section 2.2) we provide a general discussion of the channels

through which genetic distance may affect income differences, and briefly discuss these channels in

relation to the existing literature.

2.1 VTCs, Genealogical Distance and the Horizontal Transmission of Innova-

tions

2.1.1 Setup

Our model starts from the following assumptions:

a) Innovations may be transmitted vertically (across generations within a given population) and

horizontally (across different populations).

b) The horizontal diffusion of innovations is not instantaneous, but is a function of barriers to

technological and institutional diffusion.

c) Barriers to technological and institutional diffusion across societies are a function of how far

societies are from each other as a result of divergent historical paths.

6

Productive knowledge is summarized by a positive real number Ait. We assume a linear tech-

nology Yit = AitLit, where Lit is the size of the population, which implies that income per capita

is given by yit ≡ Yit/Lit = Ait.

For simplicity, we summarize all other relevant characteristics of a society (cultural habits

and traditions, language, etc.) as a point on the real line. That is, we will say that at each

time t a population i will have cultural characteristics qit, where qit is a real number.12 These

characteristics are transmitted across generations with variations.13 Over time, characteristics

change (vocabulary and grammar are modified, some cultural habits and norms are dropped while

new ones are introduced, etc.). Hence, at time t+1 a population i will have different characteristics,

given by:

qit+1 = qit + ηit+1 (1)

where qit are the characteristics inherited from the previous generations, while ηit+1 denotes cultural

change.

By the same token, the dynamics of productive knowledge includes vertical transmission across

generations as well as changes (innovations), that is:

Ait+1 = Ait +∆it+1 (2)

where ∆it+1 denotes change in productivity due to technological and institutional innovations.

Changes may take place because of original discovery by agents that belong to population i and/or

because of successful imitation/adaptation of innovations that were discovered elsewhere. The

diffusion of technological and institutional innovations can be viewed as a special case of cultural

transmission.

We are interested in the long-run process of vertical and horizontal transmission of innovations

across populations at different genealogical (i.e., genetic) distances from each other - that is, with

12Of course, this is a highly simplified and reductive way of capturing cultural differences. In general, culture is a

highly elusive and multi-faceted concept. In a well-known survey over fifty years ago Kroeber and Kluckhohn (1952)

listed 164 definitions of culture proposed by historians and social scientists. See also Boyd and Richerson (1985).

13A note on semantics is in order: while we call these characteristics "cultural" for illustrative purposes, qi could be

easily reinterpreted to include also genetically transmitted characteristics. The key points are that those characteris-

tics a) must be passed with variation from one generation to the next, and b) must affect the probability of adopting

innovations from populations with different characteristics.

7

different distances from their last common ancestor.14 To capture these relationships in the simplest

possible way, we will assume the following intergenerational structure. At time 0, there exists only

one population, with cultural characteristics q0 (normalized to zero) and productive knowledge

A0.15 At time 1 the population splits in two distinct populations (population 1 and population

2). At time 2, population 1 splits in two populations (populations 1.1 and 1.2), and population 2

splits in two populations (populations 2.1 and 2.2). This structure provides us with the minimum

number of splits we need to have variation in genealogical distances between populations at time

2. We can measure genetic distance between populations by the number of genealogical steps one

must take to reach the closest common ancestor population. Let d(i, j) denote the genetic distance

between populations i and j. Populations 1.1 and 1.2 have to go back only one step to find their

common ancestor (population 1), while populations 1.1 and 2.1 have to go back two steps to find

their common ancestor (population 0), as illustrated in Figure 1. Therefore, we have:

d(1.1, 1.2) = d(2.1, 2.2) = 1 (3)

and:

d(1.1, 2.1) = d(1.1, 2.2) = d(1.2, 2.1) = d(1.2, 2.2) = 2 (4)

What is the relationship between genealogical distance, cultural change and technological change?

In order to explore these issues, it is useful to consider the following benchmark assumptions:

A1. At each time t two populations i and j with Ait = Ajt face an identical probability πt of

discovering an original innovation that would increase productive knowledge by ∆t.

This assumption means that cultural characteristics qit per se do not have a direct effect on the

rate of technological progress: two populations with different cultural characteristics but identical

levels of productive knowledge face identical probabilities of expanding the technological frontier.

14By genealogical distance we mean the number of generations that separate two agents (in our case, populations)

from their last common ancestor. Two sisters are at a genealogical distance equal to one. Two first cousins are at a

genealogical distance equal to two. Geneticists do not express genetic distances in terms of the number of generations

back to a common ancestor, but use measures of genetic similarity across populations that are closely correlated with

the number of genealogical steps. More details on the definition and construction of genetic distance measures are

provided in Section 3.

15 In this analysis we will abstract from differences in size across populations, and assume that all populations have

identical size. For a discussion of the relationship between size and productivity, see Alesina, Spolaore and Wacziarg

(2005).

8

Figure 1: Population Tree.

In other words, cultural characteristics are assumed to be neutral with respect to the process of

innovation.16

By contrast, it is reasonable to assume that the process of imitating somebody else’s innovation

is a function of the cultural distance between the innovator and the imitator. That is, we assume:

A2. If an innovation is introduced by some population i with cultural characteristics qi, the

extent to which a population j, with cultural characteristics qj , can increase its own technological

knowledge through the imitation and adaptation of population i’s innovation will depend on the

"cultural distance" between the two populations, that is, on |qj − qi|.

But how do different populations end up with differing cultural characteristics? For the purposes

of this analysis, we will consider a simple model of cultural divergence ("mimetic drift"):17

A3. Cultural transmission follows a random walk, in which cultural characteristics are trans-

mitted vertically across populations, while "cultural change" is white noise.

16We will relax this assumption below.

17 If we reinterpret the variable qi as including some genetic characteristics, we can reinterpret the random process

as a case of "mimetic plus genetic drift". In this model we refer to qi as a purely cultural variable to stress that

VTCs include cultural features, not just genetic ones.

9

Clearly, this is a highly stylized approximation of more complex phenomena, but it does provide

a simple way to capture the dynamics of changes in neutral cultural characteristics. Specifically,

we will assume that for each population i cultural characteristics are given by:

qi = qi0 + ηi (5)

where qi0 are the characteristics of the closest ancestor (population 0 for populations 1 and 2,

population 1 for populations 1.1 and 1.2, population 2 for populations 2.1 and 2.2), and ηi is equal

to η > 0 with probability 1/2 and −η with probability 1/2.18

When cultural characteristics follow the above process, we can immediately show that on average

cultural distance between two populations is increasing in their genealogical distance. Specifically,

in our example, the expected cultural distance between populations at a genealogical distance

d(i, j) = 1 is:

E|qj − qi| | d(i, j) = 1 = η (6)

while populations at a genealogical distance d(i, j) = 2 have twice the expected cultural distance:

E|qj − qi| | d(i, j) = 2 = 2η (7)

The above relationships imply that, on average, populations that are closer genealogically will also

be closer culturally:

E|qj − qi| | d(i, j) = 2−E|qj − qi| | d(i, j) = 1 = η > 0 (8)

This is not a deterministic relationship: it is possible that two populations that are genealogically

distant will end up with more similar cultures than two populations which are more closely related.

But that outcome is less likely to be observed than the opposite. In summary, we have:

Result 1

On average, greater genealogical (i.e., genetic) distance is associated with greater cultural dis-

tance.

18We assume this process to keep the algebra as simple as possible, without loss of generality. The key assumption

is that ηi must be white noise.

10

2.1.2 Innovations and Diffusion with Neutral VTCs

We are now ready to study the relationship between diffusion of innovations, cultural change,and

genetic distance within our framework.

First of all, consider the case in which inter-population barriers to the horizontal diffusion

of innovations are prohibitive. In other terms, consider the case in which there is no horizontal

transmission of innovations, but just vertical transmission. To fix ideas, suppose that at time

t = 1, each of the two existing populations (1 and 2) could independently increase its inherited

productivity A0 by ∆ > 0 with probability π. Assuming that no other innovation takes place at

time 2, what are the expected differences in income across populations at time 2?

Populations with the same closest ancestor will inherit the same productive knowledge (either

A0 or A0 +∆) and will not differ in income per capita. That is:

E|yj − yi| | d(i, j) = 1 = 0 (9)

On the other hand, populations 1 and 2 will transfer different technologies to their descendants if

and only if one of the two population has successfully innovated at time 1 while the other population

has not. This event takes place with probability 2π(1 − π). Hence, expected income differences

across populations with genealogical distance equal to 2 are given by:

E|yj − yi| | d(i, j) = 2 = 2π(1− π)∆ (10)

Not surprisingly, when technological innovations diffuses only via vertical transmission, income

differences are strongly correlated with genealogical distance:

E|yj − yi| | d(i, j) = 2−E|yj − yi| | d(i, j) = 1 = 2π(1− π)∆ > 0 (11)

The relationship is stronger the higher is the variance of innovations across population (which, in

our example, is measured by π(1− π)), and it is highest at π = 1/2.

By contrast, if there were no barriers to the horizontal transmission of innovations across

populations, all societies would have the same income per capita independently of their genealogical

distance.19 In general, genealogical distance matters for income differences if and only if there

are barriers to horizontal diffusion. Let us consider the case in which barriers are positive but

19For all populations at time 1 and 2, we would have y = A0+ ∆ with probability 1− (1− π)2 and y = A0 with

probability (1− π)2.

11

not prohibitive. That is, if population i innovates and increases its productivity to A+ ∆ while

population j does not innovate, population j will be able to increase its total factor productivity

to:

A+max(1− β|qj − qi|)∆, 0 (12)

where β measures barriers to the diffusion of innovations across populations with different VTCs

qi and qj .20 Now, populations 1 and 2 will not end up with the same technology if and only if a)

only one of the two populations finds the innovation (an event with probability 2π(1− π)), and b)

the two populations are culturally different - that is, one experienced a cultural change equal to η

while the other experienced −η (an event with probability 1/2). If both a) and b) hold (an event

with probability π(1−π)), one of the two populations will have productivity equal to A0+∆ while

the other will have productivity equal to A0+(1− 2βη)∆. If no additional diffusion can take place

at time 2 (that is, if horizontal transmission is possible only for contemporaneous innovations), we

have:

E|yj − yi| | d(i, j) = 2−E|yj − yi| | d(i, j) = 1 = 2π(1− π)βη∆ > 0 (13)

The above equation shows that income differences are increasing in genealogical distance if and

only if there are positive barriers to diffusion (β 6= 0) and populations diverge culturally over time

( η 6= 0).

In the above example we have assumed that horizontal diffusion of the innovation introduced

at time 1 takes place only contemporaneously - that is, at time 1. The analysis can be extended to

allow for further horizontal transmission at time 2.

Consider the case in which at time t two populations (say, 1.1 and 1.2) have "inherited" tech-

nology A0 + ∆ by vertical transmission while the other two populations (say, 2.1 and 2.2) have

inherited A0 + (1 − 2βη)∆. From population 2.1’s perspective, the unadopted innovation from

period 1 is:

[A0 +∆]− [A0 + (1− 2βη)∆] = 2βη∆ (14)

If we consider this situation as equivalent to the case in which populations 1.1 and 1.2 come up

with a new innovation of size 2βη∆, we can model the adoption of that innovation by population

2.1 as:

∆2.1 = [1− β|min|q2.1 − q1.1|, |q2.1 − q1.2|]2ηβ∆ (15)

20 In the rest of the analysis, for simplicty and without loss of generality, we assume that β|qj − qi| < 1.

12

where the expression min|q2.1− q1.1|, |q2.1− q1.2| captures the fact that population 2.1 will adopt

the innovation from the population that is culturally closer.

In this case, the expected income gap between populations at different genealogical distance is

given by:

E|yj − yi| | d(i, j) = 2−E|yj − yi| | d(i, j) = 1 = π(1− π)β2η2∆ > 0 (16)

which, again, implies a positive correlation between differences in income per capita and genealogical

distance, as long as β 6= 0 and η 6= 0.21

We can summarize the above analysis as:

Result 2

Income differences across populations are increasing in genealogical (i.e., genetic) distance if

and only if there are positive barriers to the diffusion of innovations (β 6= 0) and populations

diverge culturally over time ( η 6= 0).

2.1.3 Non-Neutral VTCs

The above results have been obtained under the assumption that VTCs are neutral - that is, they

bear no direct effect on the production function and on the process of innovation itself, but only

on the process of horizontal diffusion of innovations. The assumption can be relaxed by allowing a

direct effect of cultural characteristics on the probability of innovating.22 Specifically, assume that

population i’s probability of finding an innovation is given by:

πi = π + φqi

This means that a higher qi is associated with more innovations and a lower qi with less innovations.

The analysis above can be viewed as the special case φ = 0. Under this more general assumption

equation (13) becomes:

E|yj − yi||d(i, j) = 2−E|yj − yi| | d(i, j) = 1 = 2[π(1− π) + φ2η2]βη∆ (17)

21An analogous equation can be obtained for innovations that occur in period 2. If the four populations inherit

identical technologies from period 2 and each population can find an innovation of size ∆ in period 2 with probability

π we have E|yj − yi||d(i, j) = 2−E|yj − yi||d(i, j) = 1 = 2π2(1− π)2βη∆ > 0

22 In the Appendix we will consider extensions in which cultural characteristics may directly affect not only the

probability of innovating, but also the level of productivity once the innovation has been adopted.

13

This equation shows that the larger the direct impact of cultural characteristics on the probability

of innovating, the stronger the relationship between expected income differences and genealogical

distance, provided there are barriers to diffusion (β 6= 0) and cultural heterogeneity (η 6= 0). In

other words, a direct effect of cultural characteristics on the innovation process strengthens the

relationship between genealogical (i.e., genetic) distance and income gaps, as long as there are

barriers to the diffusion of innovations, consistent with Result 2 above.

2.2 VTCs and Income Differences: A General Taxonomy

In the analytical framework presented above, we have illustrated a simple mechanism of develop-

ment diffusion implying a positive correlation between genealogical distance (i.e. distance from the

last common ancestors) and income differences. The central feature of the framework is the link

between genealogical distance and the vertical transmission of characteristics across generations. In

our model, we showed how differences in neutral characteristics (that is, characteristics that do not

have a direct effect on productivity and innovations) can explain income differences by acting as

barriers to the diffusion of innovation across populations. We then extended the model to include

possible direct effects of VTCs on productivity. Specifically, in our framework we considered a direct

effect of different characteristics on the probability of adopting productivity-enhancing innovations.

We have seen how direct effects increase the magnitude of the correlation between genetic distance

and income differences, but are not necessary for the existence of a positive correlation: barrier

effects due to neutral VTCs are sufficient to explain a positive correlation between genealogical

distance and income differences.

In our framework we have referred to the transmission of characteristics as cultural - that is, not

directly related to the transmission of DNA from parents to children. We have done that for two

reasons. One reason is conceptual: to provide a model that clearly shows how a direct link from

DNA-transmitted characteristics to economic outcomes is not necessary for our results, as long as

the vertical transmission of cultural characteristics takes place among genetically-related individ-

uals (typically, parents and children). The second reason is substantial. Our focus is on income

differences across different populations of Homo Sapiens Sapiens, taking place over a relatively

short period in terms of genetic evolution, and we expect that over that time frame divergence in

cultural characteristics have played an important role.

However, in principle the insights from our framework can be generalized to include a broader

14

set of channels through which characteristics are vertically transmitted. In general, characteristics

can be transmitted across generations through DNA (genetic transmission, or GT - e.g. eye color)

or through pure cultural interactions (cultural transmission, or CT - e.g., a specific language).

Moreover, VTCs, whether transferred through GT or CT, may affect income differences because

of a direct (D) effects on productivity or because they constitute barriers (B) to the transmission

of innovations across populations. Hence, in general one can identify four possible combinations of

mechanisms through which VTCs may affect income differences: a GT direct effect, a GT barrier

effect, a CT direct effect, and a CT barrier effect.23 The following chart summarizes the four

possibilities.

Direct Effect (D) Barrier Effect (B)

Genetic Transmission (GT) Quadrant I Quadrant II

Cultural Transmission (CT) Quadrant III Quadrant IVTaxonomy of the effects of VTCs

For instance, VTCs affecting the trade-off between quality and quantity of children in the

theoretical framework proposed by Galor and Moav (2002) to explain the Industrial Revolution

would be examples of GT direct effects (Quadrant I). GT barrier effects (Quadrant II) could stem

from visible genetically-transmitted characteristics (say, physical appearance) that do not affect

productivity directly, but introduce barriers to the diffusion of innovations and technology by

reducing exchanges and learning across populations that perceive each other as different.24 Direct

effects of cultural characteristics have been emphasized in a vast sociological literature that goes

back at least to Max Weber.25 A recent empirical study of the relationship between cultural

values and economic outcomes that is consistent with the mechanisms of Quadrant III is provided

by Tabellini (2004). The link between cultural characteristics and barriers (Quadrant IV) is at

23 It is important to notice that these conceptual types should not be viewed as completely separable, but rather as

points on a logical continuum, which may involve a mix of them. For a recent general discussion of the interactions

between biological and cultural transmission, see Richerson and Boyd (2004). Recent results in genetics that are

consistent with complex gene-culture interactions are provided by Wang et al. (2006).

24This effect is related to recent work by Guiso, Sapienzan and Zingales (2004), who argue that differences in

physical characteristics may affect the extent of trust across populations. Visible differences across ethnic groups also

play an important role in the analysis of ethnic conflict by Caselli and Coleman (2002).

25More recent references can be found in the edited volume by Harrison and Huntington (2000).

15

the core of our basic model, while our extension to non-neutral cultural characteristics may be

interpreted as an example from Quadrant III.

It is worth pointing out that the distinction between GT and CT may be useful to fix ideas, but

is not a clear-cut dichotomy. In fact, this distinction, essentially that between nature and nurture,

may be misleading from an economic perspective, as well as from a biological perspective. Generally,

the economic effects of human characteristics are likely to result from interactions of cultural and

genetic factors, with the effects of genetic characteristics on economic outcomes changing over

space and time depending on cultural characteristics, and vice versa. To illustrate this point,

consider differences across individuals within a given population (say, the U.S.). Consider a clearly

genetic characteristic of an individual, for instance having two X chromosomes. This purely genetic

characteristic is likely to have had very different effects on a person’s income and other economic

outcomes in the year 1900 and in the year 2000, because of changes in culturally transmitted

characteristics over the century. This is a case where the impact of genes on outcomes varies with

a change in cultural characteristics.26 By the same token, one can think of the differential impact

of a given cultural characteristic (say, the habit of drinking alcohol) on individuals with different

genetic characteristics (say, genetic variation in alcohol dehydrogenase, the alcohol-metabolizing

enzyme). An example of a complex interactions in which culture affects genes is the spread of the

gene for lactose tolerance in populations that domesticated cows and goats. In the interpretation

of our empirical analysis we will not dwell much on the distinction between genetic and cultural

transmission of characteristics, but interpret genetic distance as an overall measure of differences

in the whole set of VTCs.

On the other hand, we will have more to say empirically on whether our estimates capture a

direct versus a barrier effect of VTCs. Our model so far has ignored the role of geographic barriers

in the diffusion of development. In Appendix 1, we present a variety of reduced form models,

extending our basic framework to include geographic barriers. We focus on the interaction term

between geographic barriers and barriers linked to differences in VTCs. We show that a negative

effect of this interaction term on income differences arises naturally from models in which VTCs

act as barriers. The basic intuition is simple and general: under the barriers interpretation, if

populations are far apart geographically, then it should matter less that they are also distant along

26This is a variation on an example by Alison Gopnik in her comment to the now-famous Pinker vs

Spelke debate at http://www.edge.org/discourse/science-gender.html#ag. Pinker’s response is also available at

http://www.edge.org/discourse/science-gender.html.

16

some other dimension for the flow if development-enhancing innovations. In contrast, models where

VTCs bear direct effects on income do not predict such a negative interaction term, as it is the

presence or absence of traits that determines directly the extent of the income gain, irrespective of

geographic distance. We will use these theoretical insights in our empirical work to test whether

differences in VTCs act at least partly as barriers to the diffusion of development.

3 Data and Empirical Methodology

3.1 Data

Genetic Distance Since the data on genetic distance that we use as a measure of distance

in vertically-transmitted characteristics is not commonly used in the economics literature, it is

worth spending some time describing it.27 Genetic distance measures the genetic similarity of

two populations. The basic unit of analysis is the allele, or the variant taken by a gene. By

sampling populations for specific genes, geneticists have compiled data on allele frequencies, i.e.

the proportion of the population with a gene of a specific variant.28 Differences in allele frequencies

are the basis for computing summary measures of distance based on aggregated differences in allele

frequencies across various genes (or loci on a chromosome). Following Cavalli-Sforza et al. (1994),

we will use measures of FST distance, also known as coancestor coefficients (Reynolds et al., 1983).

FST distances, like most measures of genetic diversity, are based on indices of heterozygosity, the

probability that two genes at a given locus, selected at random from the relevant populations,

will be different. The construction of FST distances can be illustrated for the simple case of two

populations (a and b) of equal size, one locus, and two alleles (1 and 2). Let pa and qa be the

gene frequency of allele 1 and allele 2, respectively, in population a.29 The probability that two

randomly selected genes at a given locus are identical within the population ("homozygosity") is

27As far as we know, this is the first study of the relationship between genetic distance and differences in income

per capita across countries. Guiso, Sapienza and Zingales (2004), in a parallel study, use genetic distance between

European populations as an instrument for a measure of trust in order to explain bilateral trade flows. This is quite

different from our application, as we are interested in explaining income differences, not trade flows. Their results

are consistent with our interpretation of genetic distance as related to barriers. An economic application of measures

of genetic distance across different species is provided by Weitzman (1992).

28Allele frequencies for various genes and for most populations in the world can be conveniently searched online at

http://alfred.med.yale.edu/

29Therefore we have pa + qa = 1 and (pa + qa)2 = p2a + q2a + 2paqa = 1.

17

p2a + q2a, and the probability that they are different ("heterozygosity") is:

ha = 1− p2a + q2a = 2paqa (18)

By the same token, heterozygosity in population b is:

hb = 1− p2b + q2b = 2pbqb (19)

where pb and qb be the gene frequency of allele 1 and allele 2, respectively, in population b. The

average gene frequencies of allele 1 and 2 in the two populations are, respectively:

p =pa + pb2

(20)

and:

q =qa + qb2

(21)

Heterozygosity in the sum of the two populations is:

h = 1− p2 + q2 = 2pq (22)

By contrast, average heterozygosity is measured by:

hm =ha + hb2

(23)

FST measures the variation in the gene frequencies of populations by comparing h and hm:

FST = 1−hmh= 1− paqa + pbqb

2pq(24)

If the two populations have identical allele frequencies (pa = pb), FST is zero. On the other hand,

if the two populations are completely different at the given locus (pa = 1 and pb = 0, or pa = 0 and

pb = 1) , FST takes value 1. In general, the highest the variation in the allele frequencies across

the two populations, the higher is their FST distance. The formula can be extended to account for

L alleles, S populations, different population sizes, and to adjust for sampling bias. The details of

these generalizations are provided in Cavalli-Sforza et al. (1994, pp. 26-27).30

These measures of genetic distance have been devised mainly to reconstruct phylogenies (or

family trees) of human populations. FST (which is also known as the coancestor coefficient) can

be interpreted as the distance to the most recent common ancestors of two populations. Thus, in

30For a general discussion of measures of genetic distances, see also Nei (1987).

18

effect, genetic distance is related to how long two populations have been isolated from each other.31

If two populations split apart as the result of outmigration, their genes start to change as a result

of genetic drift (randomness) and natural selection. When calculating genetic distances in order to

study population history and phylogenesis, geneticists concentrate on neutral characteristics that

are not affected by strong directional selection occurring only in some populations and environments

(Cavalli-Sforza et al., 1994, p. 36).32 In other words, the term "neutral markers" refers to genes

affected only by drift, not natural selection.

If the populations are separated, this process of change will take them in different directions,

raising the genetic distance between them. The longer the period for which the separation lasts,

the greater will genetic distance become. More specifically, the rate of evolution is the amount of

evolutionary change, measured as genetic distance between an ancestor and a descendant, divided

by the time in which it occurred. If drift rates are constant, genetic distance can be used as a

molecular clock - that is, the time elapsed since the separation of two populations can be measured

by the genetic distance between them. Figure 2, from Cavalli-Sforza et al. (1994), illustrates the

process through which different human populations have split apart over time. Heuristically, genetic

distance between two populations is captured by the horizontal distance separating them from the

next common node in the tree.

In this paper we will use FST distance as a measure of "genealogical distance" between popula-

tions. Consistent with our theoretical framework, we expect a larger FST distance to reflect a longer

separation between populations, and hence, on average, a larger difference in VTCs. The data itself

is from Cavalli-Sforza et al. (1994), p. 75-76: we focus on the set of 42 world populations for which

they report all bilateral distances, based on 120 alleles.33 These populations are aggregated from

31 Isolation here refers to the bulk of the genetic heritage of a given population. Small amounts of interbreeding

between members of different populations do not change the big picture.

32The classic reference for the neutral theory of molecular evolution is Kimura (1968). For more details on the

neutral theory, themolecular clock hypothesis, and the construction and interpretation of measures of genetic distance,

a recent reference is Jobling et al. (2004). The fact that genetic distance is calculated with respect to neutral genetic

markets implies that genetic distance can provide an especially useful measure for the channels of Quadrants III and

IV in Figure 2 (Section 2.3).

33Cavalli-Sforza et al. (1994) also provide a different measure of genetic distance (Nei’s distance). Nei’s distance,

like FST , measures differences in allele frequencies across a set of specific genes between two populations. FST and

Nei’s distance have slightly different theoretical properties, but the differences are unimportant in practice as they

are very highly correlated, and the choice of measures does not impact our results (as we show below).

19

Figure 2: Genetic distance among 42 populations. Source: Cavalli-Sforza et al., 1994.

20

subpopulations characterized by a high level of genetic similarity. However, measures of bilateral

distance among these subpopulations are available only regionally, not for the world as a whole.34

Among the set of 42 world populations, the greatest genetic distance observed is between Mbuti

Pygmies and Papua New-Guineans, where the FST distance is 0.4573, and the smallest is between

the Danish and the English, for which the genetic distance is 0.0021.35 The mean genetic distance

among the 861 available pairs is 0.1338.

Genetic distance data is available at the population level, not at the country level. It was thus

necessary to match populations to countries. We did so using ethnic composition data from Alesina

et al. (2003). In many cases, it was possible to match ethnic group labels with population labels

from Cavalli-Sforza et al. (1994). The was supplemented with information from Encyclopedia

Britannica when the mapping of populations to countries was not achievable from ethnic group

data. Obviously, many countries feature several ethnic groups. We matched populations to the

dominant ethnic group, i.e. the one with the largest share of the country’s population.36 Random

error in the matching of populations to countries should lead us to understate the correlation

between genetic distance and income differences, and we further discuss the issue of measurement

error stemming from possible mismatches below.

The ethnic composition in Alesina et al. (2003) refers to the 1990s. This is potentially en-

dogenous with respect to current income differences if the latter are persistent and if areas with

high income potential tended to attract European immigration since 1500. This would be the case

for example under the view that the Europeans settled in North America because of a favorable

geographical environment.37 In order to construct genetic distance between countries as of 1500

34 In our empirical work, we make use of the more detailed data for Europe in order to extend our results.

35Among the more disaggregated data for Europe which we also gathered, the smallest genetic distance (equal

to 0.0009) is between the Dutch and the Danish, and the largest (equal to 0.0667) is between the Lapp and the

Sardinians. The mean genetic distance across European populations is 0.013. As can be seen, genetic distances are

roughly ten times smaller on average across populations of Europe than in the World dataset. However, we still find

that they significantly predict intra-Europe income differences.

36We have also computed measures of weighted genetic distance by using the data on each ethnic group within

a country. The measure reflects the expected distance between two randomly selected individuals, one from each

country in a pair. Weighted genetic distances are very highly correlated with unweighted ones, so as we show below

for practical purposes it does not make a big difference which one we use.

37 In fact, income differences are not very persistent at a long time horizon such as this - see Acemoglu et al. (2002).

Our own data shows that pairwise log income differences in 1500 are uncorrelated with the 1995 series in the common

21

(in an effort to obtain a variable that is more exogenous than current genetic distance), we also

mapped populations to countries using their ethnic composition as of 1500, i.e. prior to the major

colonizations of modern times. Thus, for instance, while the United States is classified as predom-

inantly populated with English people for the current match, it is classified as being populated

with North Amerindians for the 1500 match. This distinction affected mostly countries that were

colonized by Europeans since 1500 to the point where the main ethnic group is now of European

descent (New Zealand, Australia, North America and some countries in Latin America). Genetic

distance in 1500 can be used as a convenient instrument for current genetic distance. The matching

of countries to populations for 1500 is also more straightforward, possibly reducing measurement

error.

Geographic Distance. In addition to genetic distance, we also used several measures of geo-

graphic distance. The first is a measure of the greater circle (geodesic) distance between the major

cities of the countries in our sample. This comes from a new dataset compiled by researchers at

CEPII.38 This dataset features various measures of distance (between major cities, between capi-

tals, weighted using several distances between several major cities, etc.), all of which bear pairwise

correlations that exceed 99%. The dataset includes other useful controls such as whether pairs of

countries share the same primary or official language, whether the countries are contiguous, whether

they had a common colonizer, etc. We used some of these controls in our regressions.

The second measure of geographic distance that we use is latitudinal distance - i.e. simply

the absolute value of the difference in latitude between the two countries in each pair: GLAij =

|latitudei − latitudej |. Latitude could be associated with climactic factors that affect income levels

directly, as in Gallup, Mellinger and Sachs (1998) and Sachs (2001). Latitude differences would also

act as barriers to technological diffusion: Diamond (1997) suggests that barriers to the transmission

of technology are greater along the latitude direction than along the longitude direction, because

similar longitudes share the same climate, availability of domesticable animal species, soil condi-

tions, etc. We should therefore expect countries at similar latitudes to also display similar levels of

income. Finally, we control for a measure of longitudinal distance, GLOij =

¯longitudei − longitudej

¯,

to capture geographic isolation along this alternative axis.

sample (Table 1).

38The data is available free of charge at http://www.cepii.fr/anglaisgraph/bdd/distances.htm.

22

Summary Statistics Table 1, Panel A displays correlations between our various measures of

distance.39 Perhaps surprisingly, these correlations are not as high as we might have expected. For

instance, the correlation between geodesic distance and FST genetic distance is only 35.4% - though

unsurprisingly it rises to 47.8% if genetic distance is measured based on populations as they were

in 1500 (the colonization era acted to weaken the link between genetic distance and geographic

distance by shuffling populations across the globe). The correlation between alternative measures

of genetic distance, on the other hand, tends to be quite high: the FST measure and the Nei measure

of genetic distance bear a correlation of 92.9%, so it should not matter very much which one we use.

Finally, our various measures of genetic distance bear positive correlations between 14% and 22%

with the absolute value of log income differences in 1995.40 Together, these correlations suggest it

may be possible to identify separately the effects of geographic and cultural barriers on the long-run

diffusion of development.

3.2 Empirical Methodology

3.2.1 Specification

Since our empirical methodology is new, we describe it in some detail. We depart from the usual

approach, which consists of regressing income levels or income growth on a set of determinants.

Instead, we consider a specification in which the absolute difference in income between pairs of

countries (or other dependent variables such as measures of institutions and human capital) is re-

gressed on measures of distance between the countries in this pair. We computed income differences

between all pairs of countries in our sample for which income data was available, i.e. 13, 861 pairs

of countries. Thus our baseline specification is:

|log yi − log yj | = β0 + β1GDij + β2G

Gij + β3G

LAij + β4G

LOij + β05Xij + εij (25)

where GDij is genetic distance, G

Gij is geodesic distance, G

LAij is latitudinal distance, GLO

ij is longi-

tudinal distance, Xij is a set of controls capturing other types of barriers and εij is a disturbance

39Panel B presents means and standard deviations of the main variables, allowing us to assess the quantitative

magnitudes of the effects estimated in the regressions that follow.

40The data on per capita income is purchasing power-parity adjusted data from the World Bank, for the year 1995.

We also used data from the Penn World Tables version 6.1 (Summers, Heston and Aten, 2002), which made little

difference in the results. We focus on the World Bank data for 1995 as this allows us to maximize the number of

countries in our sample.

23

term to be further discussed below.41

The reason our empirical specification must involve income differences rather than a single

country’s income level on the left hand side is that this makes the use of bilateral measures of

barriers possible. There is no other way to quantify the impact of barriers, which are inherently

of a bilateral nature, on income differences. Conceptually, therefore, we depart in a major way

from existing methodologies: our regression is not directional, in the sense that the right-hand side

variable takes on the same value for each country in the pair, i.e. our specification is not simply

obtained by differencing levels regressions across pairs of countries.42

We should stress that equation (25) is a reduced form. That is, differences in income are

presumably the result of differences in institutions, technologies, human capital, savings rates, etc.,

all of which are possibly endogenous with respect to income differences. Whether income differences

are caused by these factors is the subject of a vast literature but is not primarily the subject of this

paper. This paper is concerned with barriers to the diffusion of these more proximate causes of

income differences: barriers work to explain differences in income presumably because they affect

the adoption of technologies, norms for human and physical capital accumulation, the adoption

of institutions conducive to differential economic performance, etc.. In Section 5 we will relate

barriers to differences in human capital, institutions, investment rates, population growth rates

and openness. Evaluating the role of genetic distance in affecting income differences through these

various channels, however, does not form the core of our argument.

3.2.2 Estimation

We estimate equation (25) using a new methodology. In principle, if one is willing to assume

that the measures of barriers are exogenous, equation (25) can be estimated using least squares.

However, in this case usual methods of inference will be problematic. Consider three countries,

1, 2 and 3. Observations on the dependent variable |log y1 − log y2| and |log y1 − log y3| will be

correlated by virtue of the presence of country 1 in both observations. Conditioning on the right-

hand side variables (which are bilateral in nature) should reduce cross-sectional dependence in the

41We also estimated an alternative specification where the distance measures were all entered in logs. This did not

lead to appreciable differences in the economic magnitude or statistical significance of any of the estimates.

42This, obviously, would result in adding no new information relative to the levels regressions themselves. Our

methodology is more akin to gravity regressions in the empirical trade literature than to levels or growth regressions

in the literature on comparative development.

24

errors ε12 and ε13, but we are unwilling to assume that observations on the dependent variable are

independent conditional on the regressors.43 In other words, simple least squares standard errors

will lead to misleading inferences due to spatial correlation.

Before proceeding, we note the following observations and conventions: with N countries, there

are N(N−1)/2 distinct pairs. Denote the observation on absolute value income differences between

country i and country j as dyij . Pairs are ordered so that country 1 appears in position i and is

matched with all countries from 2...N appearing in position j. Then country 2 is in position i and

is matched with 3...N appearing in position j, and so on. The last observation has country N − 1

in position i and country N in position j. We denote the non-zero off-diagonal elements of the

residual covariance matrix by σm where m is the country common to each pair.

A simple example when the number of countries is N = 4 is illustrative. In this case, under our

maintained assumption that the error covariances among pairs containing a common country m

are equal to a common value σm, the covariance matrix of the vector of residuals ε is of the form:

Ω = cov

⎛⎜⎜⎜⎜⎜⎜⎜⎜⎜⎜⎜⎜⎝

ε12

ε13

ε14

ε23

ε24

ε34

⎞⎟⎟⎟⎟⎟⎟⎟⎟⎟⎟⎟⎟⎠=

⎛⎜⎜⎜⎜⎜⎜⎜⎜⎜⎜⎜⎜⎝

σ2ε

σ1 σ2ε

σ1 σ1 σ2ε

σ2 σ3 0 σ2ε

σ2 0 σ4 σ2 σ2ε

0 σ3 σ4 σ3 σ4 σ2ε

⎞⎟⎟⎟⎟⎟⎟⎟⎟⎟⎟⎟⎟⎠In this context, controlling for a common-country fixed effect should account for the correlated part

of the error term. For this we rely on well-known results cited in Case (1991), showing that fixed

effects soak up spatial correlation, though in a context quite different from ours: we do not have

longitudinal data, and the panel nature of our dataset comes from the fact that each country is

43Another feature that reduces the dependence across pairs is the fact that the dependent variable involves the

absolute value of log income differences. Simple simulations show that under i.i.d. Normal income draws with moments

equal to those observed in our sample (Table 1b), the correlation between absolute value differences in income for

any two pair containing the same country will be about 0.22. Without taking absolute values, it is straightforward

to notice that the correlation would be exactly 0.5.

25

paired with all the other countries in the dataset.44 Following this observation, we model:

εij =NXk=1

γkδk + νij (26)

where δk = 1 if k = i or k = j, δk = 0 otherwise, and νij is a well-behaved disturbance term.

We treat δk as fixed effects, i.e. we introduce in the regression a set of N dummy variables δk

each taking on a value of one N − 1 times.45 Given our estimator, the effect of the right-hand side

variables is identified off the variation within each country, across the countries with which it is

matched.

It is in principle possible to test for the presence of spatial correlation both in the model with

common country fixed effect and without. Such tests, known as Moran’s I-tests, require specifying a

neighborhood matrix along which non-zero correlations are allowed. In our case, the neighborhood

matrix is easy to conceptualize: its entries are 1 whenever there is a common country in a pair, zero

otherwise. Unfortunately, the dimensionality of the neighborhood matrix the square of the number

of observations. In the case of our worldwide dataset, there would be 192, 127, 321 entries, making

the problem computationally intractable. We have run I-tests for our smaller dataset of European

countries, where the matrix has 87, 616 entries.46 Without dummies, Moran’s I test suggested the

presence of spatial correlation. With common country dummies added to the specification, it did

not. We find this to be encouraging.

To summarize, for each country we create a dummy variable equal to 1 if the country appears

in a given pair. We then include the full set of N dummy variables in the regression. The in-

clusion of these fixed effects soaks up the spatial correlation in the error term resulting from the

presence of each country multiple times in various country pairs. In addition, our standard errors

44Note that simply treating εi and εj as fixed effects, by including corresponding dummy variables in the regression,

will not fully address our concern. This is because, with the exception of country 1 and country N , all countries will

appear either in position i or in position j in different observations, inducing spatial correlation between these pairs.

In the example above, for instance, country 2 appears in position i in observation 1, and in position j in observation

4, inducing spatial dependence between ε12 and ε23. Simple country fixed effects would not soak up this dependence.

45The inclusion of fixed effects did not greatly alter the signs or magnitudes of the estimates of the slope coefficients

on our variables of interest, compared to simple OLS estimation. In contrast, in line with our expectations, our

common-country fixed effects technique resulted in standard errors that were quite different from (and generally

much larger than) the (wrong) ones obtained with simple OLS.

46The specification being tested is the baseline specification of Column 6 in Table 6. Details of these tests are

available upon request.

26

are heteroskedasticity-consistent (i.e. we correct standard errors to account for the fact that the

diagonal elements of Ω might differ).47

4 Empirical Results

4.1 Baseline Results

The baseline estimates of equation (25) are presented in Table 2. Columns (1) through (5) feature

our measures of distance entered one by one. The results are in line with expectations: greater

distance, whether genetic, latitudinal, longitudinal or geodesic, is significantly associated with

greater income differences. Univariate results using FST genetic distance (column 4) suggest that

a one standard deviation increase in genetic distance is associated with a 0.198 increase in log

income differences - 21.77% of this variable’s standard deviation.48 Columns (5) and (6) shows

that it matters little whether we use Nei genetic distance rather than FST genetic distance. In fact,

the impact of a standard deviation difference in Nei genetic distance is slightly larger than that of

FST distance. We will focus on FST genetic distance for the remainder of this paper, since as we

discussed it has a clear interpretation as genealogical distance.

Column (6) and (7) enters all three measures of distance together, for the Nei and FST distance

measures respectively. Interesting results emerge. First, the magnitude of the coefficient on geodesic

distance falls by over one third, suggesting that it was capturing at least in part the effect of genetic

distance. Second, the coefficient on latitudinal distance also becomes smaller in magnitude. Hence,

including genetic distance greatly reduces the effect of geographic barriers. Once controlling for

47There are obviously several ways to address the issue of spatial correlation in our context. An alternative we

considered would be to do feasible GLS by explicitly estimating the elements of Ω, and introducing the estimated Ω as a

weighing matrix in the second stage of the GLS procedure. While apparently straightforward, this is computationally

demanding as the dimensionality of Ω is large - in our application we have over 13, 000 country pairs with available

data on the variables of interest, and up to 167 covariance terms to estimate - so we leave this for future research.

We also pursued several bootstrapping strategies based on selecting subsamples such that the problem of spatial

correlation would not occur, generating results very similar to those we present here. Details and results from these

bootstraps are available upon request. In contrast to these alternatives, our approach is computationally easy to

implement.

48This percentage is the commonly used standardized beta coefficient, which we will use throughout as a convenient

measure of the economic magnitudes of our estimates. The standardized beta coefficient is reported in the last line

of each table.

27

genetic distance, the effect of a one standard deviation change in latitudinal distance on income

differences is almost halved, with a standardized beta of 6.64%: the income differences observed

unconditionally across latitudes (column 2) are not primarily geographical, but linked to differences

in vertically transmitted characteristics captured by genetic distance. Finally, longitudinal distance

bears a coefficient that is now negative and economically small.

Column (8) introduces a set of control variables that might proxy for different sets of barriers,

linguistic, historical and geographic (the Xij variables in equation (25)). Some of these variables,

such as the one reflecting linguistic similarity and colonial history, are perhaps less exogenous with

respect to income differences than the distance measures already considered, so results should be

interpreted cautiously. Several lessons emerge. First, the signs of the coefficients are largely as

expected. Contiguous countries tend to have more similar income levels, as do countries that were

ever a single political entity. Countries that have had a common colonizer share more similar income

levels, though the effect is small in magnitude. Unsurprisingly, income differences are greater when

countries were ever in a colonizer-colonized relationship. On the other hand, having substantial

fractions of the populations in each country speaking the same language bears a very small effect

on income proximity, and the sign is the opposite of that expected (we will return to the issue

of linguistic distance below). Second, these coefficients are estimated quite precisely. Third, and

perhaps most importantly, the inclusion of these controls does not modify the estimates on our

main variables of interest: the distance measures, in particular genetic distance, hardly change at

all. Across specifications, variation in genetic distance accounts for about 20% of the variation in

income differences.

Together, these results provide evidence consistent with the model of cultural barriers intro-

duced in Section 2: genetic distance enters with a positive, statistically significant and economically

meaningful coefficient in all specifications. Moreover, we find some weak empirical support for the

idea that income differences are less pronounced along similar latitudes than across latitudes, once

we control for genetic distance.

4.2 Extensions and Robustness

We now consider a number of extensions and robustness tests on the baseline specification in column

(8) of Table 3. We enter additional controls, make use of alternative genetic distance data, and

examine competing or complementary hypotheses on the diffusion of development.

28

The Interaction between Genetic and Geographic Distances. In column (1) of Table 3,

we assess whether the effects of genetic distance might depend on geodesic distance, by adding the

interaction between the two variables to the baseline specification. We find strong evidence of a

negative interaction effect: genetic distance matters less for income differences when countries are

far apart geographically, and the effect is statistically significant. As discussed in Section 2.2 and

Appendix 1, we can interpret this result as evidence in favor of the barrier interpretation for genetic

distance: the effect of genetic distance falls as physical distance increases, so these two types of

barriers act at least in part as substitutes for each other. There would be no reason to observe this

pattern of coefficients if genetic distance mattered only as the result of a genetic trait bearing a

direct effect on income levels. The magnitude of the overall effect of genetic distance, evaluated at

the mean of geodesic distance, is very close to the magnitudes estimated in the baseline specification

(with a standardized beta coefficient of 17.96% versus 18.23%).

Weighted Genetic Distance. As we described above, in our baseline results we used data

on genetic distance between the plurality populations of each country. Some countries, such as

the United States or Australia, are made up of sub-populations that are distinct, and for which

genetic distance data is available in the worldwide data on genetic distances between 42 populations

(for example Australia is composed of Aboriginals and populations of English descent). To improve

upon the measurement of genetic distance between countries, we made use of a measure of weighted

genetic distance. The measure is computed as follows: Assume that country 1 contains populations

i = 1...I and country 2 contains populations j = 1...J , denote by s1i the share of population i in

country 1 (similarly for country 2) and dij the genetic distance between populations i and j. The

weighted FST genetic distance between countries 1 and 2 is simply:

WFST =X

i=1...I

Xj=1...J

s1i × s2j × dij (27)

The interpretation of this measure is straightforward: it represents the expected genetic distance

between two randomly selected individuals, one from each country. Since there are few countries

composed of more than one populations for which there is data available in the worldwide dataset of

42 populations, and since variation induced by the plurality groups tend to dominate by construc-

tion, the correlation between weighted genetic distance and the genetic distance measure based on

plurality matching is very high - 93.4%. Unsurprisingly, therefore, in column (2) of Table 3 the

magnitude of the effect of WFST is only slightly larger than in our baseline specification, possibly

29

reflecting a lesser incidence of measurement error using weighted distance. In what follows we

continue to focus on the measure based on plurality match.

Possible Endogeneity of Genetic Distance. Next, we attempt to control for the possible

endogeneity of genetic distance with respect to income differences. While differences in (neutral)

allele frequencies between the populations of two countries obviously do not result causally from

income differences, migration could lead to a pattern of genetic distances today that is closely linked

to current income differences. Consider for instance the pattern of colonization of the New World

starting after 1500. Europeans tended to settle in larger numbers in the temperate climates of North

America and Oceania. If geographic factors bear a direct effect on income levels, and geographic

factors were not properly accounted for in the regressions through included control variables, then

genetic distance today could be positively related to income distance not because cultural distance

precluded the diffusion of development, but because similar populations settled in regions prone to

generating similar incomes.

To assess this possibility, column (3) of Table 3 excludes from the sample any pairs involving

one or more countries from the New World (defined as countries in North America, Latin America,

the Caribbean and Oceania), where the problem identified above is likely to be most acute. The

effect of genetic distance is now actually marginally larger than in column (8) of Table 2. The

difference in latitudes becomes three times larger, an observation to which we shall return below.

Next, we use our data on FST genetic distance as of 1500 as an instrument for current genetic

distance in column (4) of Table 3. This variable reflects genetic distance between populations as

they were before the great migrations of the modern era, and yet is highly correlated (65.8%) with

current genetic distance, so it fulfills the conditions of a valid instrument. Again, the magnitude

of the genetic distance effect is raised - in fact it is more than doubled in magnitude when 15th

century genetic distance is used as an instrument for current genetic distance: a standard deviation

change in genetic distance now accounts for 39.26% of a standard deviation in absolute log income

differences. As is usual in this type of application, the larger estimated effect may come from a

lower incidence of measurement error under IV - the matching of populations to countries is much

more straightforward for the 1500 match. The results suggest that, if anything, our baseline results

were understating the effect of genetic distance on income differences.

30

The Diamond Gap. Jared Diamond’s (1997) influential book stressed that differences in latitude

played an important role as barriers to the transfer of technological innovations in early human

history, and later in the pre-industrial era, an effect that could have persisted to this day. Our

estimates of the effect of latitudinal distance provided some evidence that this effect was still at

play: in our regressions we have found evidence that differences in latitudes help explain some of

the income differences across countries, and this effect was much larger when excluding the New

World from our sample. However, Diamond took his argument one step further, and argued that

Eurasia enjoyed major advantages in the development of agriculture and animal domestication

because a) it had the largest number of potentially domesticable plants and animals, and b) had a

predominantly East-West axis that allowed an easier and faster diffusion of domesticated species. By

contrast, differences in latitudes in the Americas and Africa created major environmental barriers

to the diffusion of species and innovations. More generally, Eurasia might have enjoyed additional

benefits in the production and transfer of technological and institutional innovations because of its

large size.49

To test and control for a Eurasian effect, we constructed a dummy variable that takes on

a value of 1 if one and only one of the countries in each pair is in Eurasia, and 0 otherwise

(the "Diamond gap").50 In order to test Diamond’s hypothesis, we added the Diamond gap to

regressions explaining income differences in 1995 (column 5) and in 1500 (column 6). For the

former regression, we restrict our sample to the Old World. It is appropriate to exclude the New

World from the sample when using 1995 incomes because Diamond’s theory is about the geographic

advantages that allowed Eurasians to settle and dominate the New World. If we were to include

the New World in a regression explaining income differences today, we would include the higher

income per capita of non-aboriginal populations who are there because of guns, germs and steel,

i.e. thanks to their ancestors’ Eurasian advantage. As expected, in the regression for 1995 income

differences, the Diamond gap enters with a positive and significant coefficient, and its inclusion

reduces (but does come close to eliminating) the effect of genetic distance. In column (6), using

1500 income differences as a dependent variable, the Diamond gap is again significant and large in

magnitude, despite the paucity of observations. This provides suggestive quantitative evidence in

favor of Diamond’s observation that the diffusion of development was faster in Eurasia. We also

49This point is stressed in Kremer (1993). See also Masters and McMillan (2001).

50For further tests providing statistical support for Diamond’s observations, see Olsson and Hibbs (2005).

31

conclude that genetic distance between populations plays an important role in explaining income

differences even when controlling for the environmental advantages and disadvantages associated

with Eurasia, so Diamond’s hypothesis on the long-term diffusion of development is complementary

to ours.

Climatic Factors. Our remaining robustness tests are concerned with attempts to better control

for geographic factors. Since genetic distance is correlated with geographic distance, if we do not

suitably control for geographic factors we might wrongly attribute to differences in VTCs part of the

income difference that is really due to geography. We start by controlling explicitly for differences

in climate between countries. Latitudinal difference may partly but not perfectly capture these

climatic differences. We include as an additional control (column 7) a measure of climatic similarity

based on 12 Koeppen-Geiger climate zones.51 Our measure is the average absolute value difference,

between two countries, in the percentage of land area in each of the 12 climate zones. Countries

have identical climates, under this measure, if they have identical shares of their land areas in the

same climates. As a simpler alternative, we control for the absolute difference in the percentage

of land areas in tropical climates, also based on Koeppen-Geiger zones (column 8). In both cases,

we find that the effect of genetic distance is raised slightly in magnitude. As expected, climatic

differences bear a positive relationship with income differences, and, also as expected, the inclusion

of the climatic difference variable reduces the magnitude of the effect of latitudinal differences.

Isolation controls. If our various measures of distance and geographic contiguity do not ade-

quately capture the extent to which countries may be geographically isolated from each other, we

may wrongly attribute to genetic distance the effect of geographic isolation. To address this possi-

bility, we tried a variety of measures of geographic isolation in addition to those already included as

controls in our baseline specification. In Table 4, column (1), we added a dummy variable taking on

a value of 1 if either country in each pair is an island, and a dummy similarly defined for landlocked

countries. Both, as expected, bear large and positive coefficients, consistent with the idea that

pairs where at least one country is an island or landlocked will be more isolated from each other,

51The 12 Koeppen-Geiger climate zones are: tropical rainforest climate (Af), monsoon variety of Af (Am), tropical

savannah climate (Aw), steppe climate (BS), desert climate(BW), mild humid climate with no dry season (Cf), mild

humid climate with a dry summer (Cs), mild humid climate with a dry winter (Cw), snowy-forest climate with a dry

winter (Dw), snowy-forest climate with a moist winter (Ds), tundra/polar ice climate (E) and highland climate (H).

The data, compiled by Gallup, Mellinger and Sachs, is available at http://www.ciesin.columbia.edu/eidata/.

32

resulting on average in greater income differences. However, the inclusion of these variables does

not affect the coefficient on our variable of interest, genetic distance.

Continent Effects. The largest genetic distances observed in our worldwide dataset occur be-

tween populations that live on different continents. One concern is that genetic distance may simply

be picking up the effect of cross-continental barriers to the diffusion of development, i.e. continent

effects. If this were the case, it would still leave open the question of how to interpret econom-

ically these continent effects, but to test explicitly for this possibility, we added to our baseline

specification a dummy that takes on a value of 1 if the two countries in a pair are on the same

continent (column 2 of Table 4) and a similarly defined set of dummies for six specific continents

(column 3). The results are as expected - with the exception of Asia and Oceania, living on the

same continent implies more similar income levels. The inclusion of continent dummies reduces

but does not eliminate the effect of genetic distance, which remains statistically significant. We

will provide further evidence on the within-continent effects of genetic distance using our European

dataset.

Cultural distance. As we have emphasized, cultural factors such as language, religion, norms

and values are all part of the set of VTCs captured by genetic distance. In principle, genetic

distance will reflect differences in the whole set of VTCs. Are there specific cultural traits that

genetic distance may capture, and that could be directly measurable? How much of the estimated

effect of genetic distance is attributable to differences in specific measurable VTCs? Columns (4)

through (6) of Table 4 include measures of linguistic distance and religious differences in an attempt

to answer this question.52

We use data on linguistic trees from Fearon (2003). Linguistic trees graphically display the

degree of relatedness of world languages. By counting the nodes separating language pairs in

linguistic trees, it is possible to construct discrete measures of distance between languages: in

Fearon’s data up to 15 nodes may separate languages (our variables are re-scaled from 0 to 1,

with 1 indicating high linguistic distance). Using data on the linguistic composition of countries,

and matching languages to countries, we can then construct indices of linguistic distance between

52 In our baseline specification we already included a measure of linguistic overlap - a dummy equal to 1 if at least

9% of the population in each country speaks the same language. This is a very crude measure of linguistic similarity,

and here we seek to improve upon it.

33

countries. We did so, as for genetic distance, in two ways: first, we computed a measure of

distance between plurality languages for each country pair. Second, we computed a measure of

weighted linguistic distance, representing the expected linguistic distance between two randomly

chosen individuals, one from each country in a pair (the formula is the same as that of equation 27).

This should improve the measurement of linguistic distance for countries where different languages

as spoken by different subpopulations. In columns (4) and (5) of Table 4, we include these two

measures of linguistic distance in turn. The effects are statistically significant and of the expected

sign, but they are small in magnitude: the standardized beta on linguistic distance is equal to

3.62% in column (4) and to 5.15% in column (5). Hence, our measures of linguistic distance can

only account for a small fraction of the variation in income differences. Moreover, their inclusion

does not affect the coefficient on genetic distance.53

Next, in column (6) we included an index of religious difference. The index is the average ab-

solute value difference between the countries in a pair in the population shares of 11 major world

religions, obtained from Barrett (2001).54 Again, the variable takes on the expected positive sign.

Its standardized beta is again small, at 7.66%. Moreover, its inclusion barely affects the coeffi-

cient on genetic distance compared to a regression obtained using the same sample (the estimated

coefficient falls from 2.077 to 2.032).

These results have several interpretations. The most straightforward interpretation is that

differences in VTCs captured by genetic distance are not primarily linguistic and religious in nature.

This opens up the question of what they are, an important question for future research. Another

interpretation is that the specific measures of linguistic and religious distance that we use are

inadequate. Consider linguistic distance. We use a discrete measure based on counting common

nodes from a linguistic tree. It is a well-known fact that linguistic trees and genetic trees are very

similar, as demonstrated in Cavalli-Sforza et al. (1994, p. 98-105). The reason is straightforward:

genes, like languages, are transmitted intergenerationally, and this insight is the basis for our

interpretation of genetic distance as capturing the full set of VTCs, both genetic and cultural. Yet

53The coefficient on genetic distance is raised compared to column (8) of Table 2, but this is due to the slightly

smaller sample of available country pairs when we include linguistic distance, not to the inclusion of the latter variable.

The coefficient on genetic distance is actually slightly smaller when we control for linguistic distance when the samples

are kept the same. This is what we would expect if genetic distance captured in part the effect of linguistic distance.

54These religions are Roman Catholics, Protestants, Orthodox Christians, Other Christians, Muslims, Hindus,

Buddhists, Other East Asian Religions, Jews, Other Religions, Non-Religious Population.

34

the correlation between our measure of genetic distance and linguistic distance is only about 25%.

This may be partly due to the fact that genetic distance is a continuous measure of distance, whereas

the number of nodes is a discrete measure of linguistic distance: populations may share few common

nodes but linguistic splits occurred recently, in which case one is overestimating distance, or they

may share lots of common nodes but the last split occurred a long time ago, in which case one is

underestimating distance. Better measures of linguistic distance based on lexicostatistical methods

are not widely available outside of Indo-European groups (see Dyen, Kruskal and Black, 1992).

Similarly, our measure of religious difference does not take into account the distance separating

different religions but only the similarity in religious group shares. For instance, Protestants may

be "closer" to Catholics than to Buddhists, a feature that is not be captured by our measure.

Whatever the interpretation, our results show that genetic distance is an overall measure of

differences in VTCs that is robustly correlated with income differences, and we were not able to

identify specific measures of cultural distance that could alter this conclusion.

4.3 Alternative Time Periods and Samples

Historical Income Data. Table 5 examines whether the pattern of coefficients uncovered for

income differences in 1995 held for earlier periods in history. We used income per capita data since

1500 from Maddison (2003), and repeated our basic reduced form regression for 1500, 1700, 1820,

1870, 1913 and 1960.55 For the 1500 and 1700 regressions, we use the early match for genetic

distance, i.e. genetic distance between populations as they were in 1492, prior to the discovery of

the Americas and the great migrations of modern times.56 For the subsequent periods we use the

current match. Table 5 shows that across periods, the coefficient on genetic distance is statistically

significant and positive. Moreover, the magnitudes are much larger than for the current period:

considering regressions obtained from a common sample of 26 countries (275 pairs) for which data is

55The data on income for 1960 is from the Penn World Tables version 6.1. For comparison, column (7) reproduces

the results for 1995.

56Regressions for these early periods feature at most 29 countries. These countries are Australia, Austria, Bel-

gium, Brazil, Canada, China, Denmark, Egypt, Finland, France, Germany, Greece, India, Indonesia, Ireland, Italy,

Japan, Korea, Mexico, Morocco, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, United

Kingdom, United States. There were 275 pairs (26 countries) with available data for 1500 income, and 328 pairs (29

countries) for 1700. A noteworthy feature of this sample is that it contains no countries in Sub-Saharan Africa.

35

Figure 3: Time Path of the Effect of Genetic Distance, 1500-1995.

continuously available, standardized beta coefficients range from 34.94% (1960) to 88.75% (1870).57

Thus, genetic distance is strongly positively correlated with income differences throughout modern

history. It is worth noting that genetic distance bears a large, positive and significant effect on

income differences for the past five centuries, even though income differences in 1500 and in 1995

are basically uncorrelated (the correlation between income differences in 1500 and in 1995 is −0.051

for the 325 country pairs for which data is available). This fact is consistent with our interpretation

of genetic distance as a barrier to the diffusion of innovations across populations.

Strong evidence for our model of barriers to diffusion can be obtained from plotting the time path

of coefficients, as is done in Figure 3. The result is striking. The slope coefficient of genetic distance

decreases gradually from 1500 to 1820, then spikes up in 1870 during the Industrial Revolution and

declines thereafter. This is fully consistent with our model where a major innovation (the Industrial

Revolution) initially results in large income discrepancies. These discrepancies persist in proportion

to genealogical relatedness. As more and more countries adopt the major innovation, the impact

of genetic distance progressively declines.

Overall, the historical permanence of the effect of genetic distance on income disparities over the

57Similar orders of magnitudes are obtained from unrestricted samples though magnitudes were generally lower.

36

past five centuries and the specific time path of the effect in relation to the timing of the Industrial

Revolution paint an empirical picture that is very consistent with the model of Section 2.

Genetic Distance across European Countries. Cavalli-Sforza et al. (1994) provide data on

genetic distances within certain regions of the world, in particular Europe. These data are more

disaggregated (i.e. cover more distinct populations) than the matrix of distances for 42 worldwide

populations. For Europe, they present a distance matrix for 26 populations that can be matched

readily to 25 European countries.58 Analyzing these data can be informative for several reasons.

First, it constitutes a robustness check on the worldwide results. Second, matching populations to

countries is much more straightforward for Europe than for the rest of the world, because the choice

of sampled European populations happens to match nation state boundaries. This should reduce

the incidence of measurement error. Third, genetic distances are orders of magnitude smaller across

countries of Europe, and genetic specificities within Europe have developed over the last couple of

thousand years (and not tens of thousands of years). It is very unlikely that any genetic traits have

risen to prominence within Europe as the result of strong natural selection over such a short period

of time, so a finding that genetic distance based on neutral markers within Europe is associated

with income differences would be evidence in favor of the cultural interpretation rather than a

genetic effect.

Table 6 presents the results. Much to our surprise, genetic distance is again positively and

significantly associated with income differences. Moreover, while genetic distance across European

countries are smaller than in the World sample, so are the income differences to be explained. Thus,

we find that the effect of FST distance is large in magnitude. The baseline estimate in column (6) of

Table 5, which includes several controls (we now exclude the colonial variables for obvious reasons)

suggests that a one standard deviation change in the log of genetic distance accounts for 49.18% of

the variation in log income differences. Thus, cultural barriers captured by genetic distance seem

very strongly associated with income differences. Physical distance measures correspondingly bear

small or insignificant coefficients, suggesting that geographic barriers are not a big hindrance to

the diffusion of income across countries of Europe. The last column of Table 6 shows that genetic

distance accounts for an even greater amount (79.59%) of the variation in income differences in

58These countries are Austria, Belgium, Croatia, Czech Republic, Denmark, Finland, France, Germany, Greece,

Hungary, Iceland, Ireland, Italy, Macedonia, Netherlands, Norway, Poland, Portugal, Russian Federation, Slovak

Republic, Slovenia, Spain, Sweden, Switzerland and the United Kingdom.

37

1870, during the Industrial Revolution. Again, we take these differences in estimated magnitudes

as consistent with the barrier interpretation.

4.4 Effects on the Proximate Determinants of Income Levels

Our approach to quantifying the barriers to the diffusion of development has been of a resolutely

reduced-form nature. That is, we did not specify what economic factors make incomes similar or

different, and have instead focused on the effects of geographic and cultural barriers on income

differences directly. However, differences in income result from more proximate causes. Several

prime candidates have been offered to explain differences in income per capita. These factors are

summarized in the model of Section 2 by parameter A, and we now discuss them in greater detail.

In the tradition of the Solow model, steady-state income per capita is positively affected by rates

of factor accumulation (in physical and human capital), and negatively affected by the depreciation

of capital per worker, which is more rapid when population growth is faster. The level of total factor

productivity, in growth accounting or income accounting exercises, has been found to account for

much of the variation in growth and income levels.59 What causes differences in the levels of total

factor productivity, however, is largely unknown. On a general level, TFP is "technology", though

the deeper determinants of the adoption of better technologies are generally left unspecified. A

recent literature has stressed the importance of institutions as a determinant of productivity (the

seminal contribution here is Acemoglu, Johnson and Robinson, 2001), and Glaeser et al. (2004)

have recently reemphasized the importance of human capital accumulation, rather than institutions,

as a central determinant of income levels. Finally, a large literature emphasized the role of market

size and openness as a driver of growth and income levels. We consider these five proximate causes

of income levels and examine the effects of our measures of distance on pairwise differences in the

rate of physical capital accumulation, the rate of population growth, institutional quality, the stock

of human capital and openness. These are meant to reflect the prime candidate explanations for

59For an excellent survey on this point, see Caselli (2005).

38

differences in income levels. Our empirical model can be summarized as:

Genetic distance

Geodesic distance

Latitudinal distance

Contiguity, colonial past...

⎫⎪⎪⎪⎪⎪⎪⎬⎪⎪⎪⎪⎪⎪⎭⇒

Institutional quality differences

Human capital differences

Investment differences

Population growth differences

Openness differences

⎫⎪⎪⎪⎪⎪⎪⎪⎪⎪⎬⎪⎪⎪⎪⎪⎪⎪⎪⎪⎭⇒ Income Differences

Our paper is primarily concerned with the first arrow in this diagram, and there are voluminous

literatures and unresolved debates on the respective roles of the proximate determinants of income.

Thus, we do not attempt to decompose income differences into differences into the underlying

proximate cause, a task that is both a tall order and beyond the scope of this paper. We focus

instead on investigating empirically the role of our distance measures as determinants of differences

in the proximate determinants of income.

Institutions. Engerman and Sokoloff (1997) and Acemoglu, Johnson and Robinson (2001) sug-

gest that the prime cause of economic development is the quality of a country’s institutions as the

outcome of a long-term historical process. If this is the case, countries that are distant in terms of

institutional quality should also be distant in terms of income per capita. We use a commonly-used

measure of institutional quality, the risk of expropriation variable (this variable, which ranges from

0 to 10, was used for instance in Acemoglu et al., 2001). Calculating the absolute value of the

pairwise difference in the risk of expropriation for 1990, we regressed this variable on our various

measures of geographic and cultural distance. Table 7, column (1) presents the results. The main

measures of distance are significant when entered individually, but the log of geodesic distance be-

comes small in magnitude when genetic distance and latitudinal distance are entered alongside it.60

This is interesting, as latitudinal distance had a small effect in the income differences regression.

Our finding that latitudinal distance matters for institutional differences is consistent with the

view that geographic and climatic factors have historically constituted hindrances to the diffusion

of institutions conducive to higher incomes. Our estimates suggest that a one standard deviation

change in genetic distance accounts for 17.99% of the variation in the difference in expropriation

risk.

60Results for the specifications where distance measures are entered individually are available upon request.

39

Human Capital. In a recent paper taking issue with the literature on the primacy of institutions,

Glaeser et al. (2004) suggest that variation in human capital is the fundamental cause of income

differences across countries. To examine the role of geographic and cultural barriers in preventing

countries from adopting high levels of human capital, we reran the specification of equation (25),

replacing the left hand side variable with the absolute difference in the stock of human capital,

measured by the average number of years of primary, secondary and tertiary schooling in the

population aged 25 and above in 1990.61 The results are presented in Table 7, column 2. The

pattern of coefficients is similar to that obtained for institutions. A one standard deviation change

in genetic distance accounts for 23.01% of a standard deviation in human capital differences.

Population Growth. In column (3) of Table 7, we examine the determinants of absolute dif-

ferences in rates of population growth. In neoclassical growth models, rapid population growth

reduces the steady-state level of income per worker. Thus, differences in population growth (in

turn resulting mainly from differences in mortality and fertility) are thought to be associated with

differences in income. Again, measures of cultural and geographic distance may affect how differ-

ently countries’ populations grow.62 We particularly expect geographic distance measures to be

correlated with differences in population growth, as countries located closer to the equator tend

to have higher rates of population growth. Indeed, Table 7, column (3) shows that latitudinal

distance is positively related to population growth differences (where population growth is defined

over the 1960-1990 period): the standardized beta for latitudinal distance is equal to 15%. Genetic

distance again appears to be significantly related to the dependent variable, with a standardized

beta of 12.88%. One interpretation of this finding is that differences in VTCs are associated with

persistent differences in norms of behavior affecting population growth, particularly fertility which

dominates the cross-country variation in population growth rates.63

Physical Capital Investment. The rate of investment in physical capital is also a determinant

of steady-state income levels in the neoclassical model. How do the geographic and cultural barriers

61The human capital data is from Barro and Lee (2000). Again, this is a commonly used, if imperfect, measure of

the stock of human capital.

62For a theoretical analysis of the relationships among VTCs, fertility and economic growth in the long run see

Galor and Moav (2002).

63A regression relating differences in fertility rates to our distance measures reveals a very significant and large

effect of genetic distance on fertility differences. These results are available upon request.

40

to the diffusion of development relate to differences in this proximate cause of income levels? Table

7, column (4) provides the corresponding estimates, using the Penn World Tables version 6.1 series

on the investment share of GDP for 1990. The dependent variable is the absolute difference in

these shares across country pairs. Distance measures bear positive signs when they are entered

separately as well as jointly. Genetic distance is positively related to differences in investment

rates, and its magnitude is economically significant (the standardized beta is equal to 12.52%). We

can interpret this finding as indicating that differences in VTCs hinder the adoption of norms of

investment behavior that are possibly conducive to superior economic outcomes.

Extent of the Market. Finally, there exists a very large literature on the causal links between

international openness, the extent of the market, and economic development. Recent contributions

that have emphasized the extent of the market as an important determinant of economic perfor-

mance include Ades and Glaeser (1999), Alesina, Spolaore and Wacziarg (2000, 2005), Spolaore

and Wacziarg (2005). We examined whether differences in market size are related to our various

measures of distance. Our measure of market size is the conventionally used ratio of imports plus

exports to GDP (in 1990), which proxies for access to world markets. Table 7, column (5) displays

the results when using absolute differences in this openness variable as the dependent variable.

Differences in openness seem unrelated to most of the variables included in our model, with the

exception of the post-war common colonizer variable (this could capture an Africa effect) and the

common language variable. Genetic distance appears unrelated to differences in trade openness.

To summarize, genetic distance is significantly positively associated with differences in four of

the five proximate determinants of development that we considered, and these effects are generally

large in magnitude. This confirms the results found using income differences directly, and provides

suggestive evidence that genetic distance may act through several channels.

5 Conclusion

In this paper we have documented the following facts: First, differences in income per capita

across countries are positively correlated with measures of genetic distance between populations.

Second, genetic distance, an overall measure of differences in vertically transmitted characteristics

across populations, bears an effect on income differences even when a large set of geographical and

other variables are controlled for. Third, the patterns of relationships between income differences

41

and measures of genetic and geographical distances hold not only for current worldwide data but

also for estimates of income per capita and genetic distance since 1500, as well as in a sample of

European countries. Finally, similar patterns hold when the dependent variable is differences in

human capital, institutional quality, population growth and investment rates.

These results strongly suggest that characteristics transmitted from parents to children over long

historical spans play a key role in the process of development. In particular, the results are consistent

with the view that the diffusion of technology, institutions and norms of behavior conducive to

higher incomes, is affected by differences in vertically transmitted characteristics associated with

genealogical relatedness: populations that are genetically far apart are more likely to differ in those

characteristics, and thus less likely to adopt each other’s innovations over time. The pattern of the

effects of genetic distance in space and time, and the interaction with geographical distance, suggest

that genetic distance is associated with important barriers to the diffusion of development Some

evidence, particularly the results for European countries, also suggests that these differences may

stem in substantial part from cultural (rather than purely genetic) transmission of characteristics

across generations.

A final consideration is about policy implications. A common concern with research document-

ing the importance of variables like genetic distance or geography is pessimism about its policy

implications. What use is it to know that genetic distance explains income differences, if one can-

not change genetic distance, at least in the short-run? These concerns miss a bigger point: available

policy variables may have a major impact not on genetic distance itself, but on the coefficient that

measure the effect of genetic distance on income differences. That coefficient has been changing

over time, and can change further. If we are correct in interpreting our results as evidence for long-

term barriers across different populations and cultures, significant reductions in income disparities

could be obtained by encouraging policies that reduce those barriers, including efforts to translate

and adapt technological and institutional innovations into different cultures and traditions, and

to foster cross-cultural exchanges. More work is needed - at the micro as well as macro level - in

order to understand the specific mechanisms, market forces, and policies that could facilitate the

diffusion of development across countries with distinct long-term histories and cultures.

42

References

Acemoglu, Daron, Simon Johnson and James A. Robinson (2001), "The Colonial Origins of

Comparative Development: An Empirical Investigation", American Economic Review, vol. 91, pp.

1369-1401, December.

Acemoglu, Daron, Simon Johnson and James A. Robinson (2002), “Reversal of Fortune: Ge-

ography and Institutions in the Making of the Modern World Income Distribution", Quarterly

Journal of Economics, volume 117, pp. 1231-1294, November.

Ades, Alberto and Edward Glaeser (1999), “Evidence on Growth, Increasing Returns and the

Extent of the Market”, Quarterly Journal of Economics, vol. 114, no. 3, August, pp. 1025-1045.

Alcalá, Francisco and Antonio Ciccone (2004), “Trade and Productivity”, Quarterly Journal of

Economics, vol. 119, no. 2, May.

Alesina, Alberto, Arnaud Devleeschauwer, William Easterly, Sergio Kurlat and RomainWacziarg

(2003), “Fractionalization”, Journal of Economic Growth, vol. 8, no. 2, June 2003, pp. 155-194.

Alesina, Alberto, Enrico Spolaore and Romain Wacziarg (2000), "Economic Integration and

Political Disintegration", American Economic Review, vol. 90, no. 5, pp. 1276-1296, December.

Alesina, Alberto, Enrico Spolaore and Romain Wacziarg (2004), "Trade, Growth, and the

Size of Countries", in Philippe Aghion and Steven Durlauf, eds., Handbook of Economic Growth,

Amsterdam: North Holland.

Barrett, David, George Kurian and Todd Johnson (2001), World Christian Encyclopedia (2nd

edition), New York: Oxford University Press, 2 vols.

Barro, Robert J. and Jong-Wha Lee (2000), "International Data on Educational Attainment:

Updates and Implications", Harvard CID Working Paper no. 42.

Barro, Robert J. and Xavier Sala-i-Martin (1997), "Technological Diffusion, Convergence and

Growth", Journal of Economic Growth, vol. 2, no.1, pp. 1-26.

Baumol, William (1986), "Productivity Growth, Convergence and Welfare: What the Long Run

Data Show", American Economic Review, vol. 76, pp. 1072-1085.

Bisin, Alberto and Thierry Verdier (2000), "Beyond the Melting Pot: Cultural Transmission,

Marriage, and the Evolution of Ethnic and Religious Traits," Quarterly Journal of Economics, vol.

105, no. 3, pp. 955-988.

43

Bisin, Alberto and Thierry Verdier (2001), "The Economics of Cultural Transmission and the

Evolution of Preferences," Journal of Economic Theory, vol. 97, no. 1, pp. 298-319.

Boyd, Robert and Peter J. Richerson (1985), Culture and the Evolutionary Process, Chicago:

University of Chicago Press.

Case, Anne (1991), "Spatial Patterns in Household Demand", Econometrica, vol. 59, no. 4,

July, pp. 953-965.

Caselli, Francesco (2005), "Accounting for Cross-Country Income Differences", in Philippe

Aghion and Steven Durlauf, eds., Handbook of Economic Growth, Amsterdam: North Holland.

Caselli, Francesco and John Coleman (2002), "On the Theory of Ethnic Conflict", working

paper, Harvard University.

Cavalli-Sforza, Luigi L., Paolo Menozzi and Alberto Piazza (1994), The History and Geography

of Human Genes, Princeton: Princeton University Press.

Cavalli-Sforza, Luigi L., and Francesco Cavalli-Sforza (1995), The Great Human Diasporas,

Reading, MA: Addison Wesley Publishing Company.

Cavalli-Sforza, Luigi L. and Marcus W. Feldman (1981), Cultural Transmission and Evolution,

Princeton: Princeton University Press.

Dawkins, Richard (2004), The Ancestor’s Tale: A Pilgrimage to the Dawn of Evolution, Boston,

MA: Houghton Mifflin.

Diamond, Jared (1992), The Third Chimpanzee. The Evolution and Future of the Human

Animal, New York: Harper Collins.

Diamond, Jared (1997), Guns, Germs, and Steel: The Fates of Human Societies. New York:

W. W. Norton & Company.

Dyen, Isidore, Joseph B. Kruskal and Paul Black (1992), "An Indoeuropean Classification: A

Lexicostatistical Experiment", Transactions of the American Philosophical Society, vol. 82, no. 5.

Easterly, William and Ross Levine (2003), "Tropics, Germs, and Crops: the Role of Endowments

in Economic Development", Journal of Monetary Economics, vol. 50, no. 1, January.

Engerman, Stanley L., and Kenneth L. Sokoloff (1997), "Factor Endowments, Institutions,and

Differential Paths of Growth among New World Economies: A View from Economic Historians

of the United States", in Stephen Haber, ed., How Latin America Fell Behind: Essays on the

Economic Histories of Brazil and Mexico, 1800-1914, Stanford: Stanford University Press.

44

Fearon, James (2003), "Ethnic and Cultural Diversity by Country", Journal of Economic

Growth vol. 8, no. 2 June, pp. 195-222.

Galor, Oded and Omer Moav (2002), "Natural Selection and the Origin of Economic Growth",

Quarterly Journal of Economics, vol. 117, pp. 1133-1191, November.

Galor, Oded (2005), "From Stagnation to Growth: Unified Growth Theory", in Philippe Aghion

and Steven Durlauf, eds, Handbook of Economic Growth,, Amsterdam: North-Holland.

Glaeser, Edward, Rafael LaPorta, Florencio Lopez-de-Silanes and Andrei Shleifer (2004), "Do

Institutions Cause Growth?", Journal of Economic Growth, vol. 9, pp. 271-303, September.

Guiso, Luigi, Paola Sapienza and Luigi Zingales (2004), "Cultural Biases in Economic Ex-

change", NBER Working Paper No. 11005, December.

Hall, Robert and Jones, Charles I. (1999), "Why Do Some Countries Produce so Much More

Output Per Worker than Others?", Quarterly Journal of Economics, vol. 114, no. 1, pp. 83-116,

February.

Harrison, Lawrence E. and Samuel P. Huntington, eds. (2000), Culture Matters: How Values

Shape Human Progress, New York: Basic Books.

Heston, Alan, Robert Summers and Bettina Aten (2002), "Penn World Table Version 6.1",

Center for International Comparisons at the University of Pennsylvania (CICUP), October.

Jobling, Mark A., Matthew E. Hurles and Chris Tyler-Smith (2004), Human Evolutionary

Genetics. Origins, Peoples & Diseases, New York: Garland Science.

Kimura, Motoo (1968), "Evolutionary Rate at the Molecular Level", Nature, vol. 217, pp.

624-626.

Kremer, Michael (1993), "Population Growth and Technological Change: 1,000,000 B.C. to

1990", Quarterly Journal of Economics, August, pp. 681-716.

Kroeber, A. and Kluckhohn, C. (1952), Culture. New York: Meridian Books.

Masters, William A andMargaret S. McMillan (2001), "Climate and Scale in Economic Growth,"

Journal of Economic Growth, vol. 6, no. 3, September.

Maddison, Angus (2003), The World Economy: Historical Statistics, Paris: OECDDevelopment

Center.

Gallup, John L., Andrew D. Mellinger and Jeffrey D. Sachs (1998), “Geography and Economic

Development", NBER Working Paper No. 6849.

45

Mukand, Sharun and Dani Rodrik (2005), "In Search of the Holy Grail: Policy Convergence,

Experimentation, and Economic Performance", American Economic Review, vol. 95, no. 1, pp.

374-383, March.

Nei, Masatoshi (1987), Molecular Evolutionary Genetics, New York: Columbia University Press.

Olson, Steve (2002), Mapping Human History. Discovering the Past Through Our Genes, New

York: Houghton Mifflin Company.

Olsson, Ola and Douglas A. Hibbs Jr. (2005), "Biogeography and Long-Run Economic Devel-

opment", European Economic Review, vol. 49, pp. 909-938.

Parente, Stephen L. and Edward C. Prescott (1994), "Barriers to Technology Adoption and

Development", Journal of Political Economy, vol. 102, no. 2, pp. 298-321.

Parente, Stephen L. and Edward C. Prescott (2002), Barriers to Riches, Cambridge: MIT Press.

Reynolds, J., B.S. Weir, and C.C. Cockerham (1983), "Estimation of the Coancestry Coefficient:

Basis for a Short-term Genetic Distance", Genetics, vol. 105, pp. 767-779.

Richerson Peter J. and Robert Boyd (2004), Not By Genes Alone: How Culture Transformed

Human Evolution, Chicago: University of Chicago Press.

Rodrik, Dani, Arvind Subramanian and Francesco Trebbi (2004), "Institutions Rule: The Pri-

macy of Institutions over Geography and Integration in Economic Development", Journal of Eco-

nomic Growth, vol. 9, no.2, June.

Sachs, Jeffrey (2001), "Tropical Underdevelopment", NBER Working Paper #8119.

Sachs, Jeffrey (2005), The End of Poverty: Economic Possibilities for Our Time, New York:

Penguin Books.

Spolaore, Enrico and Romain Wacziarg (2005), "Borders and Growth", Journal of Economic

Growth, vol. 10, no 4, December.

Tabellini, Guido (2005), "Culture and Institutions: Economic Development in the Regions of

Europe", IGIER Working Paper #292, June.

Wang, E.T., G. Kodama, P. Baldi, and R.K. Moyzis (2006), "Global Landscape of Recent

Inferred Darwinian Selection for Homo Sapiens", Procedeeings of the Natural Academy of Sciences,

January 3, pp.135-140.

Weitzman, Martin L. (1992), "On Diversity", Quarterly Journal of Economics, vol. 107, no. 2,

pp. 363-405, May.

46

Appendix

In this Appendix we provide some simple models in which both geographical distance and

distance in vertically transmitted characteristics (genetic distance) affect income differences by

affecting barriers to the diffusion of innovations. We will show that a negative interaction between

geographical distance and genetic distance arises naturally in those models. First we will present a

simple reduced-form specification, in which we compare barrier effects and direct effects stemming

from differences in geographical and VTCs. We will show that a negative interaction is consistent

with barrier effects, while it is not implied by direct effects. We will then sketch two specific models

of barriers, with more explicit microeconomic interpretations, and show that those models do in

fact imply a negative interaction between geographical and genetic distance. The overall message

of this Appendix is that a negative interaction between geographical distance and genetic distance

is a robust implication of models in which both distances affect income differences through their

effects on barriers to the diffusion of productivity-enhancing innovations.

Reduced-form Setup

Consider a reduced-form barrier model in which the probability that country i adopts country

j’s innovation is:

P (f, g) = P1(f)P2(g)

where f is genetic distance between i and j and g is geographical distance between i and j.

Either distance reduces the probability of adoption - i.e., it constitutes a barrier to the diffusion of

innovations: dP1/df < 0 and dP2/dg < 0. If we specify the probabilities as linear functions, we

have:

P1(f) = 1− bff

P2(g) = 1− bgg

and:

P (f, g) = 1− bff − bgg + bfbgfg

That is, the probability of adopting the same innovation is decreasing in f and g but increasing

in the interaction f × g. Hence, the expected income difference, which is inversely related to the

probability of adopting the same innovation, will be increasing in f and g, but decreasing in the

interaction f × g, as in our regressions.

47

By contrast, suppose that vertically transmitted characteristics q’s had a direct effect on total

factor productivity, with qi > qj :

TFAi = Ai = ai(qi)γ

Consistently with our framework in Section 2, assume that the difference in vertically transmitted

characteristics is a function of genetic distance. In particular, for analytical convenience and without

any loss of generality, assume:

ln qi − ln qj = Ψf

Suppose that society i comes up with an innovation that increases total factor productivity to some

level ai > aj If country j does not adopt the innovation, the difference in their levels of total factor

productivity is given by:

lnAi − lnAj = ln ai − ln aj + γ(ln qi − ln qj) = Φ+ γΨf

While if both countries adopt the innovation and achieve the same level of total factor productivity,

we have:

lnAi − lnAj = γ(ln qi − ln qj) = γΨf

By putting the two effects together (barriers effects to the adoption of the innovation, and direct

effects from vertically transmitted characteristics), we have that the expected difference in total

factor productivities between countries i and j is:

lnAi − lnAj = (1− P )Φ(ai − aj) + γΨf = [bff + bgg − bfbgfg]Φ+ γΨf = c1f + c2g + c3fg

where:

c1 = bfΦ+ γΨ > 0

c2 = bgΦ > 0

c3 = −bfbgΦ < 0

This very simple reduced-form model implies a positive effect of genetic distance and geographical

distance on income differences, and a negative interaction as long as both distances constitute

barriers to the diffusion of innovations (bf > 0 and bg > 0). It is also worth stressing that a direct

effect of genetic distance on productivity differences (γΨ > 0) increases the magnitude of c1, but

it is not necessary for c1 > 0. In other terms, barrier effects alone are sufficient for c1 > 0 and

c2 > 0, while they are necessary and sufficient for c3 < 0.

48

Two Models of Barriers

In what follows we will briefly sketch two models with a more explicit microeconomic interpre-

tation of the mechanisms through which barriers affect the adoption of innovations. Model 1 will

focus on a physical interpretation of barriers: barriers may prevent societies from observing other

societies’s innovations Model 1 considers a cost interpretation of barriers: barriers increase the

costs to adapt and imitate other societies’ innovations. In either case, the model implies a negative

interaction between geographical distance and genetic distance.

Model 1

Consider a model in which both geographical distance and genetic distance reduce the proba-

bility that a society would be able to observe another society’s innovation.

Again, consider two countries (i and j), which are separated by N geographical steps and M

cultural steps. An innovation discovered in country i must travel all N plus M steps in order to

reach country j. At each cultural step there is a probability θf that the innovation will be "lost in

translation," while at each geographical step there is a probability θg that the innovation will fail

to make it through that geographical space, where 0 < θf < 1 and 0 < θg < 1.64 Therefore the two

countries’ expected difference in productivity and income per capita will be a function of the total

probability that the innovation is lost, i.e.:

P (N,M) = 1− (1− θf )M(1− θg)

N

It is immediately apparent that this probability is a positive function of the geographical distance

(measured by N) and of the cultural distance (measured by M), and a negative function of their

interaction:∂P (N,M)

∂M= − ln(1− θf )(1− θf )

M(1− θg)N > 0

∂P (N,M)

∂N= − ln(1− θg)(1− θf )

M(1− θg)N > 0

∂2P (N,M)

∂M∂N= − ln(1− θf ) ln(1− θg)(1− θf )

M(1− θg)N < 0

Hence, these results are consistent with our reduced-form model above: differences in expected pro-

ductivity and income per capita are positively associated to measures of geographical and cultural

64Hence ln(1− θg) < 0 and ln(1− θf ) < 0.

49

distance, and negatively associated with their interaction, when those measures represent barriers

to the diffusion of productivity-enhancing innovations.

Model 2

A different mechanism that delivers analogous results is based on the assumption that geo-

graphical distance reduces the probability that the innovation is observed by the distant country,

while cultural distance increases translation costs. Suppose that country i produces an innovation

of size ∆, while country j does not. But now also assume that, when the innovation is observed in

country j, translating it into the local productive system entails a cost C which is higher for higher

cultural distance M . That is, dCdM > 0.

The innovation will be adopted in country j if and only if C(M) < ∆. Define as ΦC(M)−∆ ≥

0 the probability that the translation costs are too high, and the innovation is not adopted in

country j. SincedC

dM> 0, we have

∂Φ

∂M> 0.

Again, the two countries’ expected differences in incomes per capita will be a function of the

probability that country j does not adopt country i’s innovation. Then the probability that the two

countries have different income per capita is given by the sum of a) the probability that geographical

distance prevents country j from observing country i’s innovation, and b) the probability that

country j observes country i’s innovation but fails to adopt it because the translation costs due to

cultural distance are too high:

P (M,N) = [1− (1− θg)N ] + (1− θg)

NΦC(M)−∆ ≥ 0

Again, we have that this probability P (M,N) is increasing in N and M , while their interaction is

negative:∂P (N,M)

∂N= −[1−ΦC(M)−∆ ≥ 0] ln(1− θg)(1− θg)

N > 0

∂P (N,M)

∂M= (1− θg)

N ∂Φ

∂M> 0

∂2P (N,M)

∂M∂N= ln(1− θg)(1− θg)

N ∂Φ

∂M< 0

Hence, this model is also consistent with the reduced-form above.

50

51

Tab

le 1

– S

umm

ary

Stat

istic

s for

the

Mai

n V

aria

bles

Pane

l a. S

impl

e C

orre

latio

ns a

mon

g D

ista

nce

Mea

sure

s

Geo

desi

c di

stan

ce

Diff

. in

abso

lute

la

titud

es

Diff

. in

abso

lute

la

titud

es

F ST G

en.

Dis

t. F S

T Gen

. D

ist.,

150

0 N

ei G

en.

Dis

t. A

bs. l

og

inco

me

diff

. 199

5

Abs

. log

in

com

e di

ff. 1

500a

Diff

eren

ce in

abs

olut

e la

titud

es

0.33

11

Diff

eren

ce in

abs

olut

e lo

ngitu

des

0.84

30.

060

1

F ST G

enet

ic D

ista

nce

0.35

40.

138

0.20

51

F ST G

enet

ic D

ista

nce,

150

0 m

atch

0.

478

0.16

60.

305

0.65

8 1

N

ei G

enet

ic D

ista

nce

0.31

80.

154

0.17

10.

929

0.60

61

Abs

. log

inco

me

diff

eren

ce, 1

995

0.01

50.

104

-0.0

480.

141

0.22

60.

177

1

Abs

. log

inco

me

diff

eren

ce, 1

500a

0.15

90.

155

0.06

9-0

.096

0.

196

-0.0

86-0

.051

1A

bs. l

og in

com

e di

ffer

ence

, 170

0b 0.

141

0.24

90.

131

0.00

0 0.

072

0.07

60.

503

0.06

0(n

umbe

r of o

bser

vatio

ns: 1

3861

exc

ept a : 3

25 a

nd b : 1

431)

Pane

l b. M

eans

and

Sta

ndar

d D

evia

tions

Var

iabl

e #

Obs

. M

ean

Std.

Dev

. M

in

Max

W

orld

wid

e D

atas

et

Geo

desi

c di

stan

ce (1

000s

of k

m)

1386

17.

939

4.49

90.

010

19.9

04La

titud

inal

dis

tanc

e 13

861

0.28

30.

205

0.00

0 1.

060

Long

itudi

nal d

ista

nce

1386

10.

731

0.58

40.

000

3.50

0F S

T Gen

etic

dis

tanc

e 13

861

0.12

10.

083

0.00

0 0.

338

F ST G

enet

ic d

ista

nce,

150

0 13

861

0.12

60.

076

0.00

0 0.

356

Nei

Gen

etic

dis

tanc

e 13

861

0.02

00.

015

0.00

0 0.

062

Abs

. log

inco

me

diff

eren

ce, 1

995

1386

11.

291

0.91

30.

000

4.29

4A

bs. l

og in

com

e di

ffer

ence

, 150

0 32

50.

327

0.23

70.

000

1.01

2A

bs. l

og in

com

e di

ffer

ence

, 187

0 14

310.

647

0.48

80.

000

2.11

0E

urop

ean

Dat

aset

F S

T Gen

etic

dis

tanc

e (E

urop

e)

296

0.01

00.

006

0.00

0 0.

032

Abs

. log

inco

me

diff

eren

ce, 1

995

296

0.54

70.

431

0.00

6 1.

652

Abs

. log

inco

me

diff

eren

ce, 1

870

153

0.44

80.

304

0.00

7 1.

288

52

Tab

le 2

- B

asel

ine

regr

essi

ons

(com

mon

-cou

ntry

fixe

d ef

fect

s, de

pend

ent v

aria

ble:

abs

olut

e va

lue

of lo

g in

com

e di

ffer

ence

s, 19

95)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

G

eode

sic

Dis

tanc

e L

atitu

de

Diff

eren

ce

Lon

gitu

de

Diff

eren

ce

FST

Gen

D

ist

Nei

Gen

D

ist

All

Dis

tanc

es

(Nei

)

All

dist

ance

s (F

ST)

With

co

ntro

ls

Fst G

enet

ic D

ista

nce

2.38

8 2.

104

2.00

0

(0.1

02)*

* (0

.106

)**

(0.1

07)*

*N

ei G

enet

ic D

ista

nce

14

.893

13.4

84

(0

.569

)**

(0.5

81)*

*G

eode

sic

Dis

tanc

e

0.02

6

0.01

50.

014

0.01

3(1

000s

of k

m)

(0.0

02)*

*

(0.0

03)*

*(0

.003

)**

(0.0

03)*

*A

bsol

ute

diff

eren

ce

0.57

7

0.29

60.

294

0.23

7in

latit

udes

(0

.042

)**

(0

.047

)**

(0.0

47)*

*(0

.048

)**

Abs

olut

e di

ffer

ence

0.

081

-0

.078

-0.0

79-0

.086

in lo

ngitu

des

(0.0

14)*

*

(0.0

27)*

*(0

.027

)**

(0.0

27)*

*1

for c

ontig

uity

-0.3

88

(0

.059

)**

1 if

coun

tries

wer

e or

are

-0.2

58th

e sa

me

coun

try

(0

.071

)**

1 if

com

mon

lang

uage

0.09

3(9

% th

resh

old)

(0.0

25)*

*1

for p

airs

eve

r in

colo

nial

0.07

6re

latio

nshi

p

(0.0

77)

1 fo

r com

mon

col

oniz

er

-0

.083

post

194

5

(0.0

28)*

*1

for p

airs

cur

rent

ly in

-1.5

91co

loni

al re

latio

nshi

p

(0.3

59)*

*#

obse

rvat

ions

13

861

1386

113

861

1386

1 13

861

1386

113

861

1386

1(#

cou

ntrie

s)

(167

)(1

67)

(167

)(1

67)

(167

)(1

67)

(167

)(1

67)

Adj

uste

d R

-squ

ared

0.

750.

750.

750.

76

0.76

0.76

0.76

0.76

Effe

ct o

f 1 s.

d. c

hang

e in

bol

d re

gres

sor,

% 1

s.d.

inco

me

diff

. 12

.66%

12.9

9%5.

17%

21.7

7%

24.0

5%21

.77%

19.1

8%18

.23%

Rob

ust s

tand

ard

erro

rs in

par

enth

eses

; * si

gnifi

cant

at 1

0%; *

* si

gnifi

cant

at 5

%.

53

Tab

le 3

- R

obus

tnes

s Tes

ts a

nd E

xten

sion

s, Pa

rt I

(com

mon

-cou

ntry

fixe

d ef

fect

s, de

pend

ent v

aria

ble:

diff

eren

ce in

log

inco

me

per

capi

ta in

199

5, e

xcep

t as s

tate

d, in

col

umn

5)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

D

ist*

GD

In

tera

ctio

n W

eigh

ted

GD

W

ithou

t N

ew W

orld

IV w

ith

1500

GD

D

iam

ond

Gap

, w/o

N

ew W

orld

Inco

me

1500

, D

iam

ond

Gap

Clim

atic

di

ffer

ence

co

ntro

l

Tro

pica

l di

ffer

ence

co

ntro

l

Wei

ghte

d Fs

t Gen

etic

2.

335

D

ista

nce

(0.1

20)*

*

Fst G

enet

ic D

ista

nce

2.47

42.

306

5.33

7 1.

222

2.69

72.

872

(0

.212

)**

(0.1

73)*

*(0

.207

)**

(0.1

99)*

*(0

.129

)**

(0.1

29)*

*Fs

t Gen

etic

Dis

tanc

e,

1.

703

1500

mat

ch

(0

.441

)**

Abs

olut

e di

ffer

ence

in

0.17

20.

369

0.75

30.

244

0.72

10.

300

0.15

10.

352

Latit

udes

(0

.053

)**

(0.0

50)*

*(0

.112

)**

(0.0

49)*

* (0

.111

)**

(0.0

92)*

*(0

.068

)**

(0.0

73)*

*A

bsol

ute

diff

eren

ce in

-0

.093

0.00

3-0

.331

0.04

6 0.

013

0.00

9-0

.159

-0.1

02Lo

ngitu

des

(0.0

27)*

*(0

.028

)(0

.095

)**

(0.0

29)

(0.0

94)

(0.0

34)

(0.0

50)*

*(0

.052

)*G

eode

sic

Dis

tanc

e

0.02

30.

008

0.01

5-0

.017

-0

.032

-0.0

100.

017

0.01

0(1

000s

of k

m)

(0.0

05)*

*(0

.003

)**

(0.0

12)

(0.0

04)*

* (0

.012

)**

(0.0

08)

(0.0

06)*

*(0

.006

)D

iam

ond

Gap

0.39

00.

209

(0.0

33)*

*(0

.029

)**

Mea

sure

of c

limat

ic d

iffer

ence

0.03

7of

land

are

as, b

y 12

KG

zon

es

(0

.002

)**

Diff

eren

ce in

% la

nd a

rea

in

0.

147

KG

trop

ical

clim

ates

(0.0

22)*

*D

ista

nce

* Fs

t Gen

etic

-0

.063

D

ista

nce

(0.0

24)*

*

Obs

erva

tions

13

861

1107

976

2613

861

7626

325

1015

310

153

(# c

ount

ries)

(1

67)

(158

)(1

24)

(167

) (1

24)

(26)

(143

)(1

43)

Adj

uste

d R

-squ

ared

0.

760.

780.

75-

0.76

0.81

0.78

0.77

Effe

ct o

f 1 s.

d. c

hang

e in

bol

d re

gres

sor,

% 1

s.d.

inco

me

diff

. 17

.96%

a20

.52%

20.1

5%48

.66%

10

.68%

39.2

6%24

.13%

25.7

0%

Rob

ust s

tand

ard

erro

rs in

par

enth

eses

; * si

gnifi

cant

at 1

0%; *

* si

gnifi

cant

at 5

%. A

ll sp

ecifi

catio

ns e

xcep

t tha

t of c

olum

n 5

incl

ude

dum

mie

s equ

al

to 1

if c

ount

ries a

re c

ontig

uous

, if c

ount

ries w

ere

or a

re th

e sa

me

coun

try, i

f the

y sh

are

a co

mm

on la

ngua

ge (9

% th

resh

old)

, for

pai

rs e

ver i

n a

colo

nial

rela

tions

hip,

for p

airs

with

a c

omm

on c

olon

izer

pos

t 194

5 an

d fo

r pai

rs c

urre

ntly

in a

col

onia

l rel

atio

nshi

p, a

s in

colu

mn

(8) o

f Tab

le 2

. Th

e es

timat

ed c

oeff

icie

nts f

or th

ese

cont

rols

(not

repo

rted)

are

ava

ilabl

e up

on re

ques

t. a : e

ffec

t eva

luat

ed a

t the

mea

n of

geo

desi

c di

stan

ce.

54

Table 4 - Robustness Tests and Extensions, Part II (common-country fixed effects, dependent variable: difference in log income per capita in 1995)

(1) (2) (3) (4) (5) (6) Isolation

controls Same

continent control

Same continent controls

Linguistic distance, dominant

Linguistic distance, expected

Religious similarity

Fst Genetic Distance 2.028 1.391 0.810 2.738 2.692 2.032 (0.106)** (0.106)** (0.109)** (0.129)** (0.129)** (0.111)**Absolute difference in 0.248 0.134 -0.149 0.332 0.330 0.379Latitudes (0.047)** (0.047)** (0.054)** (0.062)** (0.062)** (0.047)**Absolute difference in -0.078 -0.127 -0.008 -0.148 -0.156 -0.040Longitudes (0.027)** (0.027)** (0.027) (0.046)** (0.046)** (0.025)Geodesic Distance 0.010 -0.007 0.004 0.007 0.007 0.002(1000s of km) (0.003)** (0.004)* (0.004) (0.006) (0.006) (0.003)=1 if either country is 0.212 an island (0.042)** =1 if either country is 0.406 landlocked (0.046)** Same Continent -0.412 Dummy (0.022)** Both in Asia 0.304 (0.037)** Both in Africa -0.960 (0.039)** Both in Europe -0.775 (0.048)** Both in North America -1.773 (0.131)** Both in Latin -0.107 America/Caribbean (0.044)** Both in Oceania 0.041 (0.163) Linguistic Distance 0.235 Index, dominant languages (0.077)** Linguistic Distance 0.469Index, Expected (0.094)**Religious Difference, 1.443based on Barrett Data (0.163)**# Observations 13861 13861 13861 10011 10004 12403(# countries) (167) (167) (167) (142) (142) (158)Adjusted R-squared 0.77 0.77 0.78 0.77 0.77 0.77Effect of 1 s.d. change in bold regressor, % 1 s.d. income diff.

18.46% 12.65% 7.37% 24.91% 24.50% 18.49%

Robust standard errors in parentheses; * significant at 10%; ** significant at 5%. All specifications include dummies equal to 1 if countries are contiguous, if countries were or are the same country, if they share a common language (9% threshold), for pairs ever in a colonial relationship, for pairs with a common colonizer post 1945 and for pairs currently in a colonial relationship, as in column (8) of Table 2. The estimated coefficients for these controls (not reported) are available upon request.

55

T

able

5 -

Reg

ress

ions

on

His

tori

cal I

ncom

e D

ata

(c

omm

on-c

ount

ry fi

xed

effe

cts,

depe

nden

t var

iabl

e: a

bsol

ute

valu

e of

log

inco

me

diff

eren

ces)

(1

) (2

) (3

) (4

) (5

) (6

) (7

)

Inco

me

1500

In

com

e 17

00

Inco

me

1820

In

com

e 18

70

Inco

me

1913

In

com

e 19

60

Inco

me

1995

Fs

t Gen

etic

Dis

tanc

e,

2.16

33.

033

15

00 m

atch

(0

.451

)**

(0.4

60)*

*

Fst G

enet

ic D

ista

nce

0.82

73.

248

3.29

41.

469

2.10

4

(0.2

31)*

*(0

.349

)**

(0.3

64)*

*(0

.105

)**

(0.1

06)*

*A

bsol

ute

diff

eren

ce

0.21

50.

547

1.03

20.

833

0.87

80.

538

0.01

4in

latit

udes

(0

.105

)**

(0.1

40)*

*(0

.068

)**

(0.1

06)*

* (0

.118

)**

(0.0

72)*

*(0

.003

)**

Abs

olut

e di

ffer

ence

-0

.096

-0.0

580.

322

0.32

7 0.

400

-0.0

050.

294

in lo

ngitu

des

(0.0

51)*

(0.0

58)

(0.0

39)*

*(0

.081

)**

(0.0

93)*

*(0

.042

)(0

.047

)**

Geo

desi

c D

ista

nce

0.

014

0.00

5-0

.035

-0.0

55

-0.0

590.

006

-0.0

79(1

000s

of k

m)

(0.0

08)*

(0.0

11)

(0.0

06)*

*(0

.011

)**

(0.0

12)*

*(0

.005

)(0

.027

)**

# O

bser

vatio

ns

325

406

1176

1711

18

9159

9513

861

(# c

ount

ries)

(2

6)(2

9)(4

9)(5

9)

(62)

(110

)(1

67)

Adj

uste

d R

-squ

ared

0.

800.

850.

840.

79

0.77

0.79

0.76

Effe

ct o

f 1 s.

d. c

hang

e in

bol

d re

gres

sor,

% 1

s.d.

inco

me

diff

. 49

.86%

46.3

5%15

.20%

40.9

3%

37.4

4%18

.04%

19.1

8%

Com

mon

sam

ple

effe

ct a

49

.86%

50.0

4%37

.67%

88.7

5%

59.5

5%34

.94%

17.6

8%

Rob

ust s

tand

ard

erro

rs in

par

enth

eses

; * si

gnifi

cant

at 1

0%; *

* si

gnifi

cant

at 5

%.

a : Eff

ect o

f 1 s.

d. c

hang

e in

bol

d re

gres

sor,

% 1

s.d.

inco

me

diff

., fo

r the

com

mon

sam

ple

of 3

25 o

bser

vatio

ns (2

6 co

untri

es),

com

pute

d to

ens

ure

com

para

bilit

y ac

ross

col

umns

.

56

Tab

le 6

- R

egre

ssio

ns fo

r E

urop

ean

Cou

ntri

es

(com

mon

cou

ntry

fixe

d ef

fect

s)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

D

ista

nce

Lat

itude

D

iffer

ence

L

ongi

tude

D

iffer

ence

FS

T G

en

Dis

t A

ll D

ista

nces

W

ith

cont

rols

18

70

Inco

me

Fst g

enet

ic d

ista

nce

in E

urop

e 35

.110

32.1

0833

.480

49.5

11

(5.8

20)*

*(7

.605

)**

(8.1

36)*

*(7

.110

)**

Geo

desi

c D

ista

nce

(100

0s o

f km

) 0.

212

0.03

30.

023

-0.1

35

(0.0

42)*

*(0

.127

)(0

.129

)(0

.134

)A

bsol

ute

diff

eren

ce in

latit

udes

0.

270

-0.8

19-1

.042

0.51

3

(0.3

65)

(0.8

68)

(0.8

75)

(1.0

29)

Abs

olut

e di

ffer

ence

in lo

ngitu

des

1.65

10.

976

0.78

91.

302

(0

.324

)**

(0.6

83)

(0.6

78)

(0.8

10)

1 fo

r con

tigui

ty

-0.0

420.

069

(0

.085

)(0

.084

)1

if co

untri

es w

ere

or a

re

-0.0

33-0

.253

the

sam

e co

untry

(0

.118

)(0

.282

)1

if co

mm

on la

ngua

ge

-0.3

23-0

.302

(9%

thre

shol

d)

(0.1

01)*

*(0

.122

)**

# ob

serv

atio

ns

296

296

296

296

296

296

171

(# c

ount

ries)

(2

5)(2

5)(2

5)(2

5)(2

5)(2

5)(1

8)A

djus

ted

R-s

quar

ed

0.77

0.74

0.77

0.78

0.79

0.80

0.80

Effe

ct o

f 1 s.

d. c

hang

e in

bol

d re

gres

sor,

% 1

s.d.

inco

me

diff

. 38

.53%

3.96

%89

.55%

51.5

7%47

.16%

49.1

8%79

.59%

Rob

ust s

tand

ard

erro

rs in

par

enth

eses

; * si

gnifi

cant

at 1

0%; *

* si

gnifi

cant

at 5

%.

57

Tab

le 7

- R

egre

ssio

ns fo

r th

e pr

oxim

ate

dete

rmin

ants

of i

ncom

e

(com

mon

-cou

ntry

fixe

d ef

fect

s, de

pend

ent v

aria

ble:

abs

olut

e va

lue

diff

eren

ce in

the

vari

able

s app

eari

ng in

the

seco

nd r

ow)

(1)

(2)

(3)

(4)

(5)

In

stitu

tions

H

uman

C

apita

l Po

pula

tion

grow

th

Inve

stm

ent

Ope

nnes

s

Fst G

enet

ic D

ista

nce

3.71

16.

259

0.33

7 10

.243

-0.0

44

(0.2

38)*

*(0

.356

)**

(0.0

29)*

* (0

.780

)**

(0.0

37)

Abs

olut

e di

ffer

ence

1.

813

2.20

10.

392

9.42

4-0

.049

in la

titud

es

(0.1

22)*

*(0

.236

)**

(0.0

22)*

* (0

.484

)**

(0.0

23)*

*A

bsol

ute

diff

eren

ce

-0.0

01-0

.059

0.06

7 0.

002

-0.0

26in

long

itude

s (0

.089

)(0

.172

)(0

.012

)**

(0.3

52)

(0.0

17)

Geo

desi

c D

ista

nce

-0

.012

-0.0

20-0

.010

-0

.000

0.00

2(1

000s

of k

m)

(0.0

11)

(0.0

21)

(0.0

02)*

* (0

.042

)(0

.002

)1

for c

ontig

uity

-0

.211

-0.3

67-0

.036

-0

.268

-0.0

27

(0.1

17)*

(0.1

77)*

*(0

.018

)**

(0.4

84)

(0.0

22)

1 if

coun

tries

wer

e or

-0

.058

-0.7

86-0

.004

-0

.910

-0.0

44ar

e th

e sa

me

coun

try

(0.1

63)

(0.2

51)*

*(0

.020

) (0

.461

)**

(0.0

32)

1 if

com

mon

lang

uage

0.

115

0.25

10.

011

0.29

9-0

.030

(9%

thre

shol

d)

(0.0

47)*

*(0

.086

)**

(0.0

07)

(0.2

02)

(0.0

10)*

*1

for p

airs

eve

r in

0.

136

-0.2

340.

054

1.26

90.

037

colo

nial

rela

tions

hip

(0.1

14)

(0.2

25)

(0.0

24)*

* (0

.612

)**

(0.0

21)*

1 fo

r com

mon

col

oniz

er

-0.1

30-0

.384

-0.0

34

-0.8

30-0

.143

post

194

5 (0

.062

)**

(0.1

02)*

*(0

.009

)**

(0.2

43)*

*(0

.015

)**

1 fo

r pai

rs c

urre

ntly

in

-0.3

05-0

.297

-0.1

93

-6.0

590.

234

colo

nial

rela

tions

hip

(0.1

92)

(1.5

41)

(0.2

61)

(0.8

45)*

*(0

.038

)**

# ob

serv

atio

ns

6328

5565

6105

89

1189

11#

coun

tries

11

310

611

1 13

413

4A

djus

ted

R-s

quar

ed

0.83

0.78

0.79

0.

800.

87Ef

fect

of 1

s.d.

cha

nge

in b

old

regr

esso

r, %

1 s.

d. o

f dep

. var

. 17

.99%

23.0

1%12

.88%

12

.52%

-0.7

4%

Rob

ust s

tand

ard

erro

rs in

par

enth

eses

; * si

gnifi

cant

at 1

0%; *

* si

gnifi

cant

at 5

%